CN116364246A - Aortic valve calcification displacement image generation method, training method and system - Google Patents
Aortic valve calcification displacement image generation method, training method and system Download PDFInfo
- Publication number
- CN116364246A CN116364246A CN202310172612.8A CN202310172612A CN116364246A CN 116364246 A CN116364246 A CN 116364246A CN 202310172612 A CN202310172612 A CN 202310172612A CN 116364246 A CN116364246 A CN 116364246A
- Authority
- CN
- China
- Prior art keywords
- image
- aortic valve
- displacement
- parameters
- calcification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000006073 displacement reaction Methods 0.000 title claims abstract description 200
- 206010050559 Aortic valve calcification Diseases 0.000 title claims abstract description 180
- 238000000034 method Methods 0.000 title claims abstract description 126
- 238000012549 training Methods 0.000 title claims abstract description 122
- 239000013598 vector Substances 0.000 claims abstract description 194
- 238000003860 storage Methods 0.000 claims abstract description 10
- 210000001765 aortic valve Anatomy 0.000 claims description 82
- 230000002980 postoperative effect Effects 0.000 claims description 71
- 230000002308 calcification Effects 0.000 claims description 43
- 238000009792 diffusion process Methods 0.000 claims description 39
- 210000000709 aorta Anatomy 0.000 claims description 38
- 230000006870 function Effects 0.000 claims description 26
- 238000012545 processing Methods 0.000 claims description 23
- 238000004590 computer program Methods 0.000 claims description 17
- 238000007781 pre-processing Methods 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims description 12
- 208000004434 Calcinosis Diseases 0.000 description 41
- 238000010968 computed tomography angiography Methods 0.000 description 33
- 230000008569 process Effects 0.000 description 25
- 238000009826 distribution Methods 0.000 description 15
- 230000001364 causal effect Effects 0.000 description 12
- 230000036772 blood pressure Effects 0.000 description 11
- 238000010586 diagram Methods 0.000 description 9
- 238000004458 analytical method Methods 0.000 description 7
- 230000010339 dilation Effects 0.000 description 6
- 230000011218 segmentation Effects 0.000 description 6
- 210000002376 aorta thoracic Anatomy 0.000 description 5
- 238000001356 surgical procedure Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000003672 processing method Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 3
- 201000010099 disease Diseases 0.000 description 3
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 3
- 208000018578 heart valve disease Diseases 0.000 description 3
- 230000002829 reductive effect Effects 0.000 description 3
- 230000002441 reversible effect Effects 0.000 description 3
- 230000024883 vasodilation Effects 0.000 description 3
- 208000024172 Cardiovascular disease Diseases 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 239000008280 blood Substances 0.000 description 2
- 210000004369 blood Anatomy 0.000 description 2
- 210000004204 blood vessel Anatomy 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 208000015532 congenital bilateral absence of vas deferens Diseases 0.000 description 2
- 238000005315 distribution function Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000036541 health Effects 0.000 description 2
- 210000003709 heart valve Anatomy 0.000 description 2
- 208000027896 Aortic valve disease Diseases 0.000 description 1
- 206010002917 Aortic valve sclerosis Diseases 0.000 description 1
- 206010008190 Cerebrovascular accident Diseases 0.000 description 1
- 208000005189 Embolism Diseases 0.000 description 1
- 206010061218 Inflammation Diseases 0.000 description 1
- 241000238633 Odonata Species 0.000 description 1
- 208000025747 Rheumatic disease Diseases 0.000 description 1
- 208000006011 Stroke Diseases 0.000 description 1
- 210000001015 abdomen Anatomy 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 210000001367 artery Anatomy 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 159000000007 calcium salts Chemical class 0.000 description 1
- 230000002490 cerebral effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000003412 degenerative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000002059 diagnostic imaging Methods 0.000 description 1
- 210000003128 head Anatomy 0.000 description 1
- 210000002837 heart atrium Anatomy 0.000 description 1
- 201000007119 infective endocarditis Diseases 0.000 description 1
- 230000004054 inflammatory process Effects 0.000 description 1
- 210000005240 left ventricle Anatomy 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 230000000670 limiting effect Effects 0.000 description 1
- 210000003141 lower extremity Anatomy 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 210000004115 mitral valve Anatomy 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001338 necrotic effect Effects 0.000 description 1
- 208000013435 necrotic lesion Diseases 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000036961 partial effect Effects 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 238000012847 principal component analysis method Methods 0.000 description 1
- 210000003102 pulmonary valve Anatomy 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000552 rheumatic effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000001225 therapeutic effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 210000000591 tricuspid valve Anatomy 0.000 description 1
Images
Classifications
-
- A—HUMAN NECESSITIES
- A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
- A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
- A61B34/00—Computer-aided surgery; Manipulators or robots specially adapted for use in surgery
- A61B34/10—Computer-aided planning, simulation or modelling of surgical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
- G16H30/20—ICT specially adapted for the handling or processing of medical images for handling medical images, e.g. DICOM, HL7 or PACS
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
- G06T2207/30048—Heart; Cardiac
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Public Health (AREA)
- Surgery (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Epidemiology (AREA)
- Heart & Thoracic Surgery (AREA)
- Biomedical Technology (AREA)
- Robotics (AREA)
- Veterinary Medicine (AREA)
- Radiology & Medical Imaging (AREA)
- Molecular Biology (AREA)
- Primary Health Care (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention discloses a generation method of aortic valve calcification displacement images, which comprises the steps of obtaining preoperative images of aortic root with valve and calcified region; obtaining a target vector, wherein the target vector is determined according to an operation scheme and/or patient sign parameters matched with a preoperative image; the preoperative image and the target vector are input into an image generation model to obtain a target image. The method can obtain accurate aortic valve calcification displacement images, and is convenient for intuitively observing the displacement of the aortic valve calcification. The invention also discloses a training method for generating the aortic valve calcification displacement image, a system, equipment and a storage medium for generating the aortic valve calcification displacement image.
Description
Technical Field
The invention relates to the field of artificial intelligence, in particular to an aortic valve calcification displacement image generation method, an aortic valve calcification displacement image training system, aortic valve calcification displacement image training equipment and a computer storage medium.
Background
In China, cardiovascular diseases are diseases with the first morbidity and mortality, and heart valve diseases are one of cardiovascular diseases, so that the life and health of people are seriously threatened. Heart valve surgery is as high as 20 tens of thousands of cases per year, which is the leading heart surgery in adults.
The heart valve is a valve positioned between an atrium and a ventricle or between the ventricle and an artery, and comprises a mitral valve, a tricuspid valve, a pulmonary valve and an aortic valve, and the opening and closing of the valve are realized by a transvalve pressure difference on two sides of valve leaflets, so that the unidirectional flow and the minimum return flow of blood are ensured.
The aortic valve is located between the left ventricle with a large pressure load and the aorta, and is liable to cause aortic valve diseases. Calcification refers to the process that a certain tissue in a human body is necrotized under the action of some factors, and then calcium salt in the body is deposited in a necrotic focus, so that the pathological changes are limited and tend to be stable. Essentially, calcification is a defensive reaction of the body to lesions that facilitates clearance of necrotic lesions and resolution of inflammation. Although most calcifications are a defensive response, some calcifications can lead to adverse consequences, aortic valve Calcification (CAVD), a type of calcified heart valve disease (Calcific heart valvedisease, CHVD). It is a disease caused by calcification of the heart aortic valve, which can cause embolism to cause cerebral apoplexy and also can cause infectious endocarditis, thus seriously endangering the health of people. Aortic valve calcification is a common and frequent disease worldwide: in developed countries, more than 20% of people over 65 years old have different degrees of aortic valve sclerosis, and in China, the valve calcification detection rate reaches 12.5% in people over 50 years old, wherein the aortic valve involvement accounts for 94.4%. The incidence of aortic valve calcification has been statistically increasing year by year.
At present, it is believed that CAVD achieves a "non-rescue" approach, and the most prominent treatment for aortic valve calcification is surgery, e.g., aortic valve replacement (Transcatheter aortic valve replacement, TAVR). At present, for aortic valve calcification, only a pre-operative CTA (CT Angiography) contrast image of the patient can be acquired, and the pre-operative image is analyzed through artificial experience and the displacement of the aortic valve calcification after the operation is predicted in an abstract way.
At present, a displacement image of the aortic valve calcification cannot be directly and accurately generated through the preoperative image of the aortic valve calcification so as to enable people to intuitively observe the displacement of the aortic valve calcification.
Disclosure of Invention
The invention aims to solve the problem that at the present stage, the displacement image of the aortic valve calcification can not be directly and accurately generated through the preoperative image of the aortic valve calcification so as to intuitively observe the displacement of the aortic valve calcification by people.
In a first aspect, the method for generating the aortic valve calcification displacement image provided by the invention is an image processing method, and in particular relates to a medical image processing method, which can directly and accurately generate the aortic valve calcification displacement image so as to enable people to intuitively observe the aortic valve calcification displacement.
In order to solve the technical problems, the embodiment of the invention discloses a method for generating calcified displacement images of an aortic valve, which is used for acquiring preoperative images of the root and calcified region of the aortic valve; obtaining a target vector, wherein the target vector is determined according to an operation scheme and/or patient sign parameters matched with a preoperative image; the preoperative image and the target vector are input into an image generation model to obtain a target image.
By adopting the technical scheme, the preoperative image of the aortic root with the valve and the calcified region directly and accurately generates the displacement image of the calcification of the aortic valve through the image generation model of the calcification displacement of the aortic valve, so that people can intuitively observe the displacement of the calcification of the aortic valve.
According to another embodiment of the present invention, obtaining the target vector includes: acquiring and preprocessing original parameters to obtain a first vector, wherein the original parameters at least comprise surgical scheme and/or patient sign parameters; the target vector is obtained by the text encoder based on the first vector.
According to another embodiment of the invention, the obtaining and preprocessing of the original parameters to obtain the first vector comprises: obtaining original parameters according to the preoperative image; extracting parameters related to aortic valve calcification displacement from the original parameters as target parameters; a first vector is constructed from the target parameters.
According to another embodiment of the present invention, extracting parameters related to calcification displacement of aortic valve from original parameters as target parameters includes: extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters, and formatting the extraction parameters to obtain target parameters.
According to another embodiment of the present invention, extracting parameters related to calcification displacement of aortic valve from original parameters as target parameters includes: extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters; weighting the extracted parameter to obtain a weighting parameter; normalizing the weighting parameters to obtain target parameters.
According to another embodiment of the invention, the image generation model of aortic valve calcification displacement employs a diffusion model.
According to another embodiment of the invention, the image generation model of aortic valve calcification displacement employs a conditional generation countermeasure network or flow model.
According to another embodiment of the present invention, before acquiring a preoperative image of a root and calcified region of an aortic valve, the image generation method comprises: acquiring a CTA contrast image of the aorta before operation; the CTA contrast image of the pre-operative aorta is segmented to obtain a pre-operative image of the valved aortic root and calcified region.
In a second aspect, embodiments of the present invention disclose an aortic valve calcification displacement image generation system comprising: the acquisition module is used for acquiring preoperative images of the aortic root with the valve and the calcified region, and acquiring target vectors, wherein the target vectors are determined according to the operation scheme and/or the patient sign parameters; the processing module is used for inputting the preoperative image and the target vector into an image generation model so as to obtain a target image; and the output module is used for outputting the target image.
By adopting the technical scheme, the image generation system for aortic valve calcification displacement can acquire preoperative images and target vectors through the acquisition module, and the image generation module processes the acquired preoperative images and target vectors to obtain target images and output the target images. The image generation system of the aortic valve calcification displacement can directly and accurately generate an aortic valve calcification displacement image (namely a target image) through the image generation module according to preoperative images of the aortic root with the valve and the calcified region and the target vector, and provide an image for intuitively observing the displacement of the aortic valve calcification for people. The target image generated by the image generation system of the aortic valve calcification displacement is accurate and is fit with the expected target image.
In a third aspect, the invention discloses a training method for generating aortic valve calcification displacement images, which comprises the following steps: acquiring preoperative images of the root and calcified region of the aortic valve; obtaining a target vector, wherein the target vector is determined according to an operation scheme and/or patient sign parameters matched with a preoperative image; processing the preoperative image and the target vector through a model to be trained to obtain a post-operative predicted image; updating the target vector of the model to be trained based on the postoperative predicted image until the model training condition is met, and obtaining an image generation model of the aortic valve calcification displacement.
By adopting the technical scheme, the training mode of combining the text with the image is used for training the model to be trained, so that the image generation model can generate the target image matched with the information content of the guide information, and the accuracy of the generated target image is improved.
According to another embodiment of the invention, the training method further comprises: a post-operative image of the root and calcified region of the valved aortic valve corresponding to the pre-operative image is acquired.
According to another embodiment of the present invention, the model training conditions include: acquiring and preprocessing original parameters to obtain a first vector, wherein the original parameters at least comprise surgical scheme and/or patient sign parameters; performing calcification region image generation training on the preoperative image; obtaining a target vector by a text encoder according to the first vector; and performing repeated iterative training on the target vector and the postoperative predicted image according to the preoperative image and the postoperative image to obtain an image generation model of the aortic valve calcification displacement, wherein the postoperative predicted image is obtained according to the preoperative image and the target vector.
According to another embodiment of the present invention, the training of the target vector and the post-operation predicted image is performed for a plurality of times according to the pre-operation image and the post-operation image to obtain an image generation model of the aortic valve calcification displacement, which comprises: during the first iterative training, the input of an image generation model of the aortic valve calcification displacement is a preoperative image and a target vector, wherein the target vector is obtained by original parameters according to the preoperative image; and in other iterative training except the first time, the input of the image generation model of the aortic valve calcification displacement is a postoperative predicted image obtained in the previous iterative training and a target vector obtained in the previous iterative training, wherein the target vector is determined according to a loss function obtained in the previous iterative training.
According to another embodiment of the present invention, the training of the target vector and the post-operation predicted image is performed for a plurality of times according to the pre-operation image and the post-operation image to obtain an image generation model of the aortic valve calcification displacement, which comprises: obtaining a first image based on the preoperative image and the target vector, and taking the first image as a post-operative predicted image; and (3) performing iteration: obtaining a loss function based on the postoperative predicted image and the postoperative image, adjusting the target vector into an update vector based on the loss function, and taking the update vector as a new target vector; obtaining a second image based on the postoperative predicted image and the new target vector, and taking the second image as the postoperative predicted image; and obtaining an image generation model of the calcification displacement of the aortic valve until the loss function is smaller than a preset value.
According to another embodiment of the invention, before acquiring a post-operative image of the root and calcified region of the valved aortic valve corresponding to the pre-operative image, the method comprises: acquiring a CTA contrast image of the aorta after operation; and dividing the CTA contrast image of the aorta after operation to obtain an image of the root of the aortic valve with calcification area after operation.
In a fourth aspect, embodiments of the present invention disclose a training system for aortic valve calcification displacement image generation, comprising: the acquisition module is used for acquiring preoperative images of the root and calcified region of the aortic valve; obtaining a target vector, wherein the target vector is determined according to an operation scheme and/or patient sign parameters matched with a preoperative image; the training module is used for processing the preoperative image and the target vector through a model to be trained to obtain a post-operative predicted image; updating the target vector of the model to be trained based on the postoperative predicted image until the model training condition is met, and obtaining an image generation model of the aortic valve calcification displacement.
By adopting the technical scheme, the image predicted by the image generation training system of the aortic valve calcification displacement is accurate and is fit with reality.
In a fifth aspect, embodiments of the present invention disclose an electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the aortic valve calcification displacement image generation method in any of the embodiments of the first aspect and/or implementing the training method in any of the embodiments of the third aspect when the computer program is executed.
In a sixth aspect, embodiments of the present invention disclose a computer readable storage medium storing a computer program which, when executed by a processor, implements the aortic valve calcification displacement image generation method in any of the embodiments of the first aspect, and/or implements the training method of aortic valve calcification displacement image generation in any of the embodiments of the third aspect.
Drawings
FIG. 1 is a flowchart I of a method for generating aortic valve calcification displacement images in accordance with an embodiment of the present invention;
FIG. 2 illustrates a second flowchart of a method of generating aortic valve calcification displacement images in an embodiment of the invention;
FIG. 3 illustrates a third flowchart of a method of aortic valve calcification displacement image generation in an embodiment of the invention;
FIG. 4 illustrates a fourth flowchart of a method of generating aortic valve calcification displacement images in an embodiment of the invention;
FIG. 5 illustrates a fifth flowchart of a method of generating aortic valve calcification displacement images in an embodiment of the invention;
FIG. 6 shows a schematic diagram of the structure of an aortic valve calcification displacement image generation system in an embodiment of the invention;
FIG. 7 shows a flowchart six of a method of aortic valve calcification displacement image generation in an embodiment of the invention;
FIG. 8 illustrates a seventh flowchart of a method of aortic valve calcification displacement image generation in an embodiment of the invention;
FIG. 9 shows a flowchart eight of a method of aortic valve calcification displacement image generation in an embodiment of the invention;
FIG. 10 shows a flowchart nine of a method of generating aortic valve calcification displacement images in an embodiment of the invention;
FIG. 11 illustrates a flowchart ten of a method of aortic valve calcification displacement image generation in an embodiment of the invention;
FIG. 12 is a schematic diagram of a training system for aortic valve calcification displacement image generation in accordance with an embodiment of the invention;
fig. 13 shows a schematic structural diagram of an electronic device for generating an aortic valve calcification displacement image in an embodiment of the invention.
Detailed Description
Further advantages and effects of the present invention will become apparent to those skilled in the art from the disclosure of the present specification, by describing the embodiments of the present invention with specific examples. While the description of the invention will be described in connection with the preferred embodiments, it is not intended to limit the inventive features to the implementation. Rather, the purpose of the invention described in connection with the embodiments is to cover other alternatives or modifications, which may be extended by the claims based on the invention. The following description will include numerous specific details in order to provide a thorough understanding of the present invention. The invention may be practiced without these specific details. Furthermore, some specific details are omitted from the description in order to avoid obscuring the invention. It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
It should be noted that in this specification, like reference numerals and letters denote like items in the following figures, and thus once an item is defined in one figure, no further definition or explanation thereof is necessary in the following figures.
The terms "first," "second," and the like are used merely to distinguish between descriptions and are not to be construed as indicating or implying relative importance.
In the description of the present embodiment, it should also be noted that, unless explicitly specified and limited otherwise, the terms "disposed," "connected," and "connected" are to be construed broadly, and may be, for example, fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; can be directly connected or indirectly connected through an intermediate medium, and can be communication between two elements. The specific meaning of the above terms in the present embodiment can be understood in a specific case by those of ordinary skill in the art.
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.
Aortic valves are very important portal structures in the heart. The aorta is the thickest, most pressurized large vessel of the whole body. Blood passes through the heart, through the aortic valve to the aorta, and then to the whole body, including the head, abdomen, and lower extremities. Therefore, the function of such a portal structure is also very important. The aortic valve, an important valve structure in the aorta, is in fact a very thin structure, and the normal aortic valve resembles the wings of a dragonfly.
The aortic valve is very flexible and easy to open and close. However, under the influence of rheumatic or senile degenerative valve changes, the valve thickens and slowly forms a stone-like change, i.e. aortic valve calcification. After calcification, the aortic valve activity is reduced. When the valve is open, the size of the opening of the aortic valve in the open and closed state will be significantly affected, as the stone-like structure following calcification of the aortic valve will hinder the valve's movement. In addition, aortic valve calcification also involves the annulus, and long valve structures may also calcifie, even the aorta. Therefore, the image generation method for aortic valve calcification displacement needs to acquire images of the aortic root, aortic valve, and calcified region, that is, the image of the aortic root and calcified region with valve mentioned below.
The aortic valve calcification displacement image generation method, the training system, the training equipment and the computer storage medium provided by the invention all have the non-therapeutic purpose.
The direct object of the aortic valve calcification displacement image generation method and the training method provided by the invention is preoperative images and surgical schemes.
The terminology involved in one or more embodiments of the present invention is explained:
Generating a model: refers to a model that can be sampled from a given data distribution.
Diffusion model, the intuition behind the diffusion model derives from physics. In physics, gas molecules diffuse from a high concentration region to a low concentration region, which is similar to information loss due to interference of noise. So by introducing noise and then attempting to generate an image by denoising. Through multiple iterations over a period of time, the model learns to generate new images each time given some noise input. The diffusion model works by learning the attenuation of information due to noise and then using the learned pattern to generate an image. One standard diffusion model has two main process domains, forward diffusion and reverse diffusion. In the forward diffusion phase, the image is contaminated with noise that is gradually introduced until the image becomes completely random noise. In the reverse process, the predictive noise is removed step by step at each time step using a series of Markov chains to recover the data from the Gaussian noise.
Diffusion process: a random process of gradually increasing noise on data and an inverse process to recover the data distribution can be used to construct the generative model.
Flow model: the depth generation model is reversible, and can realize bidirectional transformation between hidden variables and original data. In the sequence generation mode, in the process of each iteration, causal relations among a plurality of variables are sequentially generated, each generation is based on the last generation result, and the causal relations are continuously enriched along with the progress of the causal relation generation process.
Generating an antagonizing network: namely, the generated type countermeasure network is a deep learning model. The model is built up of at least two modules in a frame: the mutual game learning of the generative model G (Generative Model) and the discriminant model D (Discriminative Model) produces a fairly good output. Such as: g is a model for creating a high-resolution image, and D is a model for detecting whether or not the image is an original natural image. The objective of G is to make D determine whether the high-resolution image generated by G is an unnatural image or not, D determines whether the input image is an original natural image or an unnatural image generated by G as far as possible, and the parameters of G and D are continuously updated iteratively until the generation countermeasure network meets the convergence condition.
The generator network: for generating a high resolution image from a low resolution image. The generator may be a deep learning based convolutional neural network.
At present, the information that can provide the analysis of the calcification of the aortic valve is limited, for example, only the acquired pre-operative CTA (CT Angiography) contrast image of the patient can be referred to.
In a first aspect, the present invention provides a method for generating an aortic valve calcification displacement image, which can generate an image of the aortic valve calcification displacement according to a target vector and a preoperative image of a root and a calcified region of an aortic valve through an image generation model of the aortic valve calcification displacement, so as to allow people to intuitively observe the displacement of the calcified region of the aortic valve.
Compared to the prior art, the present application provides more information for analyzing aortic valve calcification, for example, in addition to pre-operative CTA (CT Angiography) contrast images of the patient, post-operative predictive images generated by image generation models.
The aortic valve calcification displacement image generation method provided by the first aspect of the invention is an image processing method, and particularly relates to a medical image processing method, which can directly and accurately generate the aortic valve calcification displacement image so as to enable people to intuitively observe the displacement of the aortic valve calcification.
Referring to fig. 1, in an embodiment of the present invention, there is provided a method for generating an aortic valve calcification displacement image, the method comprising:
S1: a preoperative image of the root of the valved aortic arch and the calcified region is acquired.
Preoperative images of the aortic root and calcified region with valve, i.e. images of the aortic root, aortic valve and calcified region prior to transcatheter aortic valve replacement. Hereinafter simply referred to as preoperative images.
S2: a target vector is acquired, the target vector being determined according to a surgical plan and/or patient sign parameters adapted to the preoperative image.
In this embodiment, the pre-operative image may be analyzed manually or by electronic devices such as a computer to obtain a corresponding surgical plan, and/or patient sign parameters may be collected, and the target vector may be determined according to the surgical plan and/or patient sign parameters. Illustratively, patient sign parameters include age, blood pressure, etc.; the surgical plan samples include pre-dilation pressure of the aortic vessel, surgical consumables such as aortic valve stent model, forceps, clips, etc.
S3: the preoperative image and the target vector are input into an image generation model to obtain a target image.
By adopting the technical scheme, the preoperative image and the target vector are input into the image generation model of the aortic valve calcification displacement, the preoperative image is processed through the calcification region generator in the image generation model, the surgical scheme and/or the patient sign parameters matched with the preoperative image are processed through the text encoder in the image generation model to obtain the target vector, and the target image, namely the displacement image of the aortic valve calcification is directly and accurately generated through the guidance of the image and the text, so that people can intuitively observe the displacement of the aortic valve calcification.
Referring to fig. 2 in combination with fig. 1, in some possible embodiments of the invention, S2: the obtaining of the target vector includes:
s21: the method comprises the steps of obtaining and preprocessing raw parameters to obtain a first vector, wherein the raw parameters at least comprise surgical plan and/or patient sign parameters.
In this embodiment, the raw parameters include at least surgical plan and/or patient characterization parameters. Illustratively, patient sign parameters include age, blood pressure, etc.; surgical protocols include pre-dilation pressure of aortic vessels, surgical consumables such as aortic valve stent model, forceps, clips, etc. The original parameters are empirically obtained from pre-operative images of the root of the valved aorta and calcified region.
Preprocessing the original parameters at least comprises preprocessing the original parameters at least comprising extracting parameters related to calcification displacement of the aortic valve from the original parameters, and constructing the extracted parameters into a first vector.
In this embodiment, the original parameters adapted to the preoperative image are obtained by analyzing the preoperative image, and the first vector is constructed on the basis of the original parameters, so that unification and standardization of the original parameters are realized, and the subsequent data processing is facilitated.
S22: and obtaining a target vector through a text encoder according to the first vector. A target vector of a specific format is formed so as to be input as a guide condition to the image generation model.
In this embodiment, the text encoder is used as a neural network after training, takes the first vector as input, and performs operations such as dimension increasing, autocorrelation and the like to obtain the target vector with dimension and data type meeting requirements.
Referring to fig. 3 in combination with fig. 2, in some possible embodiments provided by the present invention, S21: acquiring and preprocessing original parameters to obtain a first vector, including:
s211: the original parameters are obtained from the preoperative image. As described above, the pre-operative image is analyzed to obtain the original parameters that are adapted thereto. Either by manual analysis or by computer analysis.
S212: and extracting parameters related to the calcification displacement of the aortic valve from the original parameters as target parameters.
S213: a first vector is constructed from the target parameters.
In this embodiment, the extracted parameters are parameters related to calcification displacement of the aortic valve in the original parameters, such as blood pressure of the patient, pre-expansion pressure of the aortic blood vessel, model of the aortic valve stent, and the like, and the extracted parameters are used as target parameters and are constructed into a first vector, so that subsequent data processing is facilitated.
In some other possible embodiments provided by the present invention, S212: extracting parameters related to aortic valve calcification displacement from original parameters as target parameters, wherein the parameters comprise:
S2120: extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters, and formatting the extraction parameters to obtain target parameters. (not shown in the drawings)
Illustratively, parameters related to aortic valve calcification displacement, i.e. blood pressure, pre-dilation pressure of aortic vessels, aortic valve stent model, etc. extracted from the original parameters are subjected to a formatting process to form formatted text. Illustratively, the extracted parameters are tabulated by a formatting process. The target parameters are formatted to have a uniform format, and compared with parameters with non-uniform formats, the target parameters are simpler and more convenient to construct.
Referring to fig. 4 in combination with fig. 3, in some other possible embodiments provided by the present invention, S212: extracting parameters related to aortic valve calcification displacement from original parameters as target parameters, wherein the parameters comprise:
s2121: extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters;
s2122: weighting the extracted parameter to obtain a weighting parameter;
s2123: normalizing the weighting parameters to obtain target parameters.
In this embodiment, step S212: in the process of extracting the parameter related to the calcification displacement of the aortic valve from the original parameters as the target parameter and obtaining the target parameter by extracting the related parameter, the processing mode of the extracted related parameter is optional, and is not limited to the extraction related parameter, the weighting value and the normalization processing proposed in the embodiment.
In some possible embodiments provided by the present invention, the image generation model of aortic valve calcification displacement employs a diffusion model.
Illustratively, the image generation model of aortic valve calcification displacement employs a semantically guided diffusion model. The process of denoising by using different semantic information to guide the diffusion model is realized by changing the guide conditions. The image generation model of the aortic valve calcification displacement based on the semantic guided diffusion model does not need operations such as adjusting posterior distribution, processing a distribution function, training an additional discriminator, restricting the generation process to a normalized flow, and the like, so that the training process is more stable. Illustratively, the guiding condition of the semantic guiding diffusion model is image guiding or text guiding or guiding of a combination of image and text.
In some other possible embodiments, no-classification guides are used instead of semantic guides. And (3) without diffusion guidance of a classifier, calculating a Gaussian distribution mean value by estimating noise estimation based on a DDPM (Denoising Diffusion Probabilistic Model, denoising diffusion probability model) model, and performing noise estimation by using additional conditions, namely adding the classification guidance into the noise estimation model.
In some possible embodiments provided by the invention, the image generation model of the aortic valve calcification displacement adopts a conditional generation countermeasure network, and the conditional generation countermeasure network has mature technology, so that the implementation is simpler.
In some other possible embodiments provided by the present invention, the image generation model of aortic valve calcification displacement employs other generation models, such as a flow model. The flow model identifies the causal relationship among the variables in a sequence generation mode, and along with iterative updating of the generated flow model, the generated flow model can have the capability of generating a better causal relationship sequence, and a plurality of causal sequences are not required to be sampled to select a better causal relationship sequence, so that the calculation cost can be reduced, and the convergence rate of the model is improved. The invention is not limited to the adopted generation model.
Referring to fig. 5 in combination with fig. 1, in some possible embodiments provided by the present invention, S1: before acquiring the preoperative image of the root and calcified region of the aorta to be treated, the image generation method comprises the following steps:
s01: a CTA contrast image of the aorta prior to the operation is acquired. CTA contrast images are CT Angiography (CT Angiography). The CTA contrast image of the aorta prior to the operation is obtained, for example, by medical photography or other possible methods.
S02: the CTA contrast image of the pre-operative aorta is segmented to obtain a pre-operative image of the valved aortic root and calcified region.
Illustratively, in step S02, the preoperative image of the root of the valved aortic valve and the calcified region may be obtained by a segmentation algorithm, or may be obtained by manual segmentation; the segmentation algorithm adopts an AI technology or adopts a traditional image processing technology, and the segmentation algorithm adopted by the invention is not limited.
In this embodiment, a CTA contrast image of the aorta before operation is obtained by preprocessing in steps S01 and S02 to obtain a pre-operation image of the root and calcified region of the valved aorta required in step S1.
In a second aspect, referring to fig. 6, the present invention provides an aortic valve calcification displacement image generation system 1 comprising:
the acquisition module 11 is used for acquiring preoperative images of the aortic root with the valve and the calcified region and acquiring target vectors, wherein the target vectors are determined according to the operation scheme and/or the patient sign parameters;
a processing module 12 for inputting the preoperative image and the target vector into an image generation model to obtain a target image;
an output module 13 for outputting the target image.
By adopting the technical scheme, the preoperative image sample of the aortic root with the valve and the calcified region can directly and accurately generate the displacement image of the aortic valve calcification through the image generation system of the aortic valve calcification displacement, and the image for intuitively observing the displacement of the aortic valve calcification is provided for people. The image predicted by the image generation system of the aortic valve calcification displacement is accurate and is fit with reality.
In a third aspect, referring to fig. 7, the present invention provides a training method for generating aortic valve calcification displacement images, comprising:
s4: a preoperative image of the root of the valved aortic arch and the calcified region is acquired.
As described in the first aspect, the method for generating an aortic valve calcification displacement image provides a preoperative image of the aortic root and calcified region with valve, that is, an image of the aortic root, aortic valve and calcified region before transcatheter aortic valve replacement.
In this embodiment, the preoperative image mentioned in the training method of the image generation model is a sample image obtained from a database.
S5: a target vector is acquired, the target vector being determined according to a surgical plan and/or patient sign parameters adapted to the preoperative image.
The preoperative images are analyzed, and surgical protocols and/or acquisition of patient parameters (i.e., raw parameters) for the preoperative images are obtained through manual experience or electronic means. And (3) carrying out data processing on the original parameters to obtain target vectors so as to prepare guide conditions for the training image generation model.
S6: and processing the preoperative image and the target vector through a model to be trained to obtain a postoperative predicted image.
In this embodiment, the target vector is training input data of the model to be trained, and the post-operation predicted image is used as supervision data of the training result. The post-operation predicted image is a generated image obtained by an image generation model of aortic valve calcification displacement according to the pre-operation image and the target vector. The post-operation predicted image is obtained according to the pre-operation image and the target vector, and the correlation between the generated image and the input image and the text is enhanced by a training mode of combining the images and the text, so that the obtained post-operation predicted image is more accurate.
S7: updating the target vector of the model to be trained based on the postoperative predicted image until the model training condition is met, and obtaining an image generation model of the aortic valve calcification displacement.
By adopting the technical scheme, the image generation model of the aortic valve calcification displacement is trained by using the preoperative image, the target vector and the postoperative predictive image of the aortic root and the calcification region, so that the image generation model can generate the target image matched with the information content of the guiding information, and the accuracy of the generated target image is improved.
In some possible embodiments provided by the present invention, the training method for aortic valve calcification displacement image generation further comprises: s0: a post-operative image of the root and calcified region of the valved aortic valve corresponding to the pre-operative image is acquired. (not shown in the drawings)
In some possible embodiments, referring to fig. 8, S0: before acquiring the postoperative image of the aortic root with valve and calcified region corresponding to the preoperative image, the training method for generating the aortic valve calcification displacement image further comprises:
s03: a CTA contrast image of the aorta after the operation is acquired.
S04: and dividing the CTA contrast image of the aorta after operation to obtain an image of the root of the aortic valve with calcification area after operation.
In this embodiment, CTA contrast images of the aorta after the operation are obtained by preprocessing in step S03 and step S04 to obtain the post-operation images of the aortic root with valve and calcified region required in step S0, that is, the images of the aortic root, aortic valve and calcified region after the transcatheter aortic valve replacement operation. In this embodiment, the post-operation image mentioned in the training method of the image generation model is a sample image obtained from a database.
In some possible embodiments provided by the present invention, referring to fig. 9 in combination with fig. 7, S7: updating a target vector of a model to be trained based on a postoperative predicted image until a model training condition is met, wherein the model training condition comprises the following steps of:
S71: the method comprises the steps of obtaining and preprocessing raw parameters to obtain a first vector, wherein the raw parameters at least comprise surgical plan and/or patient sign parameters.
In this embodiment, the raw parameters include at least patient parameters and/or surgical plan. Illustratively, patient sign parameters include age, blood pressure, etc.; surgical protocols include pre-dilation pressure of aortic vessels, surgical consumables such as aortic valve stent model, forceps, clips, etc. The original parameters are empirically obtained from pre-operative images of the root of the valved aorta and calcified region.
The first vector is obtained by preprocessing the original parameters, wherein the preprocessing at least comprises the steps of extracting parameters related to calcification displacement of the aortic valve from the original parameters, and constructing the extracted parameters into the first vector. The extracted parameters are parameters related to calcification displacement of the aortic valve in the original parameters, such as blood pressure of a patient, pre-expansion pressure of an aortic blood vessel, model of an aortic valve stent and the like, and the extracted parameters are constructed into a first vector so as to facilitate subsequent data processing.
S72: and performing calcification region image generation training on the preoperative image. To reduce the randomness of the image generation.
A diffusion model, also a generative model, is used to generate data similar to training data. Basically, the working principle of the diffusion model is to destroy training data by continuously adding gaussian noise, and then learn recovery data by reversing this noise process. After training, the diffusion model may be used to introduce randomly sampled noise into the model, and the data may be generated by learning a denoising process.
Illustratively, the image generation model is trained based on a semantic guided diffusion model. In this embodiment, the calcification region generator is trained based on the semantic guided diffusion model with pre-operative images as input. The pre-operative image acquired in S1 is used as an input of a diffusion model, and a calcified region generator is trained by using a specific method. The method adds a distribution randomly selected from a preset group of Gaussian distributions to the input as a supervision target, and samples from the standard Gaussian distribution as training input of a generator to complete training.
After step S32 is completed, the calcified region is entered and the calcified region generator will obtain a result that is substantially consistent with the input to reduce randomness.
S73: the target vector is obtained by the text encoder based on the first vector.
Illustratively, the text encoder takes as input a trained multi-layer perceptron neural network, and takes as input a first vector, a target vector whose dimensions and data types meet the requirements.
S74: and performing repeated iterative training on the target vector and the postoperative predicted image according to the preoperative image and the postoperative image to obtain an image generation model of the aortic valve calcification displacement, wherein the postoperative predicted image is obtained according to the preoperative image and the target vector.
The training is performed in step S74, so that the input is a preoperative image and the output is a post-operative predicted image, i.e. the input is inconsistent with the output to increase randomness.
In some possible embodiments provided by the present invention, referring to fig. 10 in combination with fig. 7, S74: performing iterative training on the target vector and the post-operation predicted image for a plurality of times according to the pre-operation image and the post-operation image to obtain an image generation model of the aortic valve calcification displacement, wherein the image generation model comprises the following steps:
s741: during the first iterative training, the input of an image generation model of the aortic valve calcification displacement is a preoperative image and a target vector, wherein the target vector is obtained by original parameters according to the preoperative image;
s742: and in other iterative training except the first time, the input of the image generation model of the aortic valve calcification displacement is a postoperative predicted image obtained in the previous iterative training and a target vector obtained in the previous iterative training, wherein the target vector is determined according to a loss function obtained in the previous iterative training.
In some possible embodiments provided by the present invention, referring to fig. 11 in combination with fig. 7, S74: performing iterative training on the target vector and the post-operation predicted image for a plurality of times according to the pre-operation image and the post-operation image to obtain an image generation model of the aortic valve calcification displacement, wherein the image generation model comprises the following steps:
S743: obtaining a first image based on the preoperative image and the target vector, and taking the first image as a post-operative predicted image;
and (3) performing iteration: s744: obtaining a loss function based on the postoperative predicted image and the postoperative image, adjusting the target vector into an update vector based on the loss function, and taking the update vector as a new target vector;
s745: obtaining a second image based on the postoperative predicted image and the new target vector, and taking the second image as the postoperative predicted image;
s746: and obtaining an image generation model of the calcification displacement of the aortic valve until the loss function is smaller than a preset value.
Through iterative training, an image generation model of the aortic valve calcification displacement can directly and accurately generate a displacement image of the aortic valve calcification according to preoperative images of the aortic root with the valve and the calcification region and a target vector. The limiting loss function is smaller than a preset value, so that a target image generated by the image generation model obtained after training is more accurate.
Specifically, referring to fig. 11 in combination with fig. 7, S743: obtaining a first image based on the preoperative image and the target vector, and taking the first image as a post-operative predicted image;
s744: obtaining a loss function based on the postoperative predicted image and the postoperative image, and judging whether the loss function is smaller than a preset value or not;
If not, execution S745: adjusting the target vector into an update vector based on the loss function, taking the update vector as a new target vector, obtaining a second image based on the postoperative predicted image and the new target vector, and taking the second image as the postoperative predicted image;
if so, execution S746: an image generation model of the aortic valve calcification displacement is obtained.
In some possible embodiments provided by the present invention, the image generation model of aortic valve calcification displacement employs a diffusion model. Illustratively, the image generation model of aortic valve calcification displacement employs a semantically guided diffusion model. The process of denoising by using different semantic information to guide the diffusion model is realized by changing the guide conditions. The image generation model of the aortic valve calcification displacement based on the semantic guided diffusion model does not need operations such as adjusting posterior distribution, processing a distribution function, training an additional discriminator, restricting the generation process to a normalized flow, and the like, so that the training process is more stable. The guidance condition is, for example, image guidance or text guidance or guidance of a combination of image and text.
In some possible embodiments provided by the invention, since the semantic guidance based diffusion model has certain randomness in the generation process, steps S4 to S7 shown in fig. 7 are performed multiple times, and the obtained multiple predicted calcified region displacement images are superimposed to obtain a final prediction result. The results represent in the form of a heat map the possible displacement of the calcified region of the aortic valve after transcatheter aortic valve replacement under a specific surgical protocol, making the resulting predicted displacement image of the calcified region more accurate.
In some other possible embodiments, no-classification guides are used instead of semantic guides. Classifier-free diffusion guidance, based on a DDPM (Denoising Diffusion Probabilistic Model, denoising diffusion probability model) model, computes a gaussian distribution mean by estimating the noise estimate, where the noise estimate is made using additional conditions, i.e., adding classification guidance to the noise estimate model, but retraining the diffusion model every time a different guidance type is needed. The non-classification guide can avoid the step-by-step training process, has better controllability and improves the probability of successful training.
In some possible embodiments provided by the invention, the image generation model of the aortic valve calcification displacement adopts a conditional generation countermeasure network, and the conditional generation countermeasure network has mature technology, so that the implementation is simpler.
In some other possible embodiments provided by the invention, the image generation model of the aortic valve calcification displacement adopts other generation models, such as a flow model, the flow model identifies causal relations among variables in a sequence generation mode, and as the iteration of generating the flow model is updated, the flow model can have the capability of generating a better causal relation sequence without sampling a large number of causal sequences to select a better causal relation sequence from a plurality of causal sequences, so that the calculation cost can be reduced, and the convergence speed of the model is improved.
The generation model adopted by the invention can be a generation countermeasure network, a variation automatic encoder and the like.
With reference to fig. 7-11 in combination with fig. 3-5, an exemplary embodiment of a training method for aortic valve calcification displacement image generation is described below.
First, a preoperative image of the aortic root with valve and calcified region, a postoperative image of the aortic root with valve and calcified region corresponding to the preoperative image, and a target vector are acquired.
That is, a CTA contrast image of the aorta is acquired and a data set is created, and one example of data in the data set includes at least a CTA contrast image of the aorta before the operation, a CTA contrast image of the aorta after the operation, and an original parameter, wherein the CTA contrast image of the aorta before the operation refers to a pre-operation image of the aortic valve replacement by a catheter, and the CTA contrast image of the aorta after the operation refers to a post-operation image of the aortic valve replacement by a catheter. Each instance of data in the dataset is preprocessed to obtain pre-operative images of the valved aortic root and calcified region, post-operative images of the valved aortic root and calcified region corresponding to the pre-operative images, and a target vector. The pretreatment specifically comprises the following steps:
referring to fig. 5, a preoperative image of the root of the valved aortic arch and the calcified region is acquired through steps S01 and S02. Namely, S01: a CTA contrast image of the aorta prior to the operation is acquired.
S02: the CTA contrast image of the pre-operative aorta is segmented to obtain a pre-operative image of the valved aortic root and calcified region.
By executing steps S01 and S02, S4 as shown in fig. 7 is realized: a preoperative image of the root of the valved aortic arch and the calcified region is acquired.
Referring to fig. 3 in combination with fig. 9, a first vector is acquired through steps S211, S212, and S213 (S71 as shown in fig. 9: acquiring and preprocessing original parameters to obtain the first vector). That is to say,
s211: the original parameters are obtained from the preoperative image.
Illustratively, the raw parameters include at least patient sign parameters, such as age, blood pressure; and surgical protocols such as pre-dilation pressure of aortic vessels, surgical procedure consumables such as aortic valve stent model, forceps, clips, etc.
S212: extracting parameters related to aortic valve calcification displacement from the original parameters as target parameters;
illustratively, among the original parameters, the parameters related to aortic valve calcification displacement are blood pressure, pre-dilation pressure of aortic vessels, aortic valve stent model, etc.
Parameters related to aortic valve calcification displacement are extracted from the original parameters, and the basis of the extracted parameters is that numerical simulation is carried out through the displacement process of aortic valve calcification in transcatheter aortic valve replacement, and meanwhile, the original parameters are analyzed by using a principal component analysis method.
Referring to fig. 3 and 4, S212: extracting parameters related to aortic valve calcification displacement from original parameters as target parameters, wherein the parameters comprise:
s2122: and extracting parameters related to the calcification displacement of the aortic valve from the original parameters as extraction parameters.
S2123: the weighted parameters are obtained by weighting the extracted parameters, and the parameters with larger influence on the calcification displacement of the aortic valve are weighted more heavily, so that the given parameters are better influenced, and the image of the calcification displacement of the aortic valve which is predicted and generated is more accurate.
S2124: normalizing the weighting parameters to obtain target parameters.
Referring to fig. 3, S213 is performed: a first vector is obtained from the target parameter.
As described above, the preoperative image is acquired through steps S01 and S02 as shown in fig. 5, and the first vector is acquired through steps S211, 2312 and S213 as shown in fig. 3 for step S73 shown in fig. 9: the target vector is obtained by the text encoder according to the first vector, providing data preparation.
Referring to fig. 8, a post-operation image is acquired through step S03, step S04, and step S0. That is to say,
s03: acquiring a CTA contrast image of the aorta after operation;
s04: and dividing the CTA contrast image of the aorta after operation to obtain an image of the root of the aortic valve with calcification area after operation.
S0: a post-operative image of the root and calcified region of the valved aortic valve corresponding to the pre-operative image is acquired.
Next, a generator is constructed based on the semantic guided diffusion model, the generator consisting of an image generator and a text encoder. In this embodiment, the image generator is a calcified region generator, and uses the calcified region of the preoperative image or the calcified region of the postoperative image as the input of the diffusion model, adds a distribution selected randomly from a preset group of gaussian distributions to the input as a supervision target, and samples from the standard gaussian distribution as the training input of the calcified region generator to complete training.
Referring to fig. 7, step S6 is performed: and processing the preoperative image and the target vector through the model to be trained to obtain a postoperative predicted image.
In the embodiment, the post-operation predicted image is obtained according to the pre-operation image and the target vector, and the correlation between the image and the text is enhanced by combining the image with the text in a training mode, so that the obtained post-operation predicted image is more accurate.
S7: updating the target vector of the model to be trained based on the postoperative predicted image until the model training condition is met, and obtaining an image generation model of the aortic valve calcification displacement. The method specifically comprises the following steps:
Referring to fig. 9 in combination with fig. 7, S7: updating a target vector of a model to be trained based on a postoperative predicted image until a model training condition is met, wherein the model training condition comprises the following steps of:
s71: the original parameters are acquired and preprocessed to obtain the first vector, the original parameters being obtained from the preoperative image. The method for acquiring the first vector in the training method of the image generation model may be the same as the method for acquiring the first vector in the image generation method.
Illustratively, there are two methods for obtaining the first vector:
the first method is, referring to fig. 3, execution S211: the original parameters are obtained from the preoperative image. In this embodiment, the original parameters are obtained by manual analysis of the preoperative image or analysis of electronic devices such as a computer, for example, by a targeted surgical plan and/or patient sign parameters.
S212 is executed: and extracting parameters related to the calcification displacement of the aortic valve from the original parameters as target parameters. Illustratively, stent model, pre-vasodilation pressure, etc. in the surgical plan are extracted as target parameters, blood pressure, etc. in the patient's vital parameters.
S213 is executed: a first vector is obtained from the target parameter.
The second method is, referring to fig. 3, execution S211: the original parameters are obtained from the preoperative image. In this embodiment, the original parameters are obtained by manual analysis of the preoperative image or analysis of electronic devices such as a computer, for example, by a targeted surgical plan and/or patient sign parameters.
S212 is executed: and extracting parameters related to the calcification displacement of the aortic valve from the original parameters as target parameters. Illustratively, stent model, pre-vasodilation pressure, etc. in the surgical plan are extracted as target parameters, blood pressure, etc. in the patient's vital parameters.
Next, referring to fig. 4, S2121 is performed: and extracting parameters related to the calcification displacement of the aortic valve from the original parameters as extraction parameters. Illustratively, stent model, pre-vasodilation pressure, etc. in the surgical plan are extracted as target parameters, blood pressure, etc. in the patient's vital parameters.
Execution S2122: and weighting the extracted parameter to obtain a weighting parameter. And respectively weighting the sizes of the influences of the extracted parameters on the generation of the aortic valve calcification displacement image, and weighting the large parameters, otherwise, weighting the small parameters.
Execution S2123: normalizing the weighting parameters to obtain target parameters.
With continued reference to fig. 3, S213 is performed: a first vector is obtained from the target parameter.
In other embodiments of the present invention, the method for acquiring the first vector is not limited.
Referring to fig. 9, S72 is performed: the pre-operative image is trained for calcified region image generation to reduce the randomness of image generation.
In this embodiment, the calcification region generator is trained based on the semantic guided diffusion model with pre-operative images as input. That is, with the preoperative image acquired in S4 shown in fig. 7 as an input to the diffusion model, a calcified region generator is trained using a specific method. The method adds a distribution randomly selected from a preset group of Gaussian distributions to the input as a supervision target, and samples from the standard Gaussian distribution as training input of a generator to complete training.
After step S72 is completed, the calcified region is entered and the calcified region generator will obtain results that are substantially consistent with the input to reduce randomness.
With continued reference to fig. 9, S73: the target vector is obtained by the text encoder according to the first vector obtained in step S71. That is, on the basis of obtaining a calcified region generator, a first vector is constructed from the extracted parameters related to aortic valve calcification displacement, a target vector is passed through a text encoder according to the first vector, and the target vector is input as a guide condition into a semantic guide diffusion model. At each sampling of the sampling sub-process, the first vector passes through the text encoder and then serves as a target vector, and the target vector is used as a guide condition to be inserted into the generating process.
S74: and performing repeated iterative training on the target vector and the postoperative predicted image according to the preoperative image and the postoperative image to obtain an image generation model of the aortic valve calcification displacement, wherein the postoperative predicted image is obtained according to the preoperative image and the target vector. The training is performed in step S74, so that the input is a preoperative image and the output is a post-operative predicted image, i.e. the input is inconsistent with the output to increase randomness. The method comprises the following steps:
referring to fig. 10 in combination with fig. 9, S74: performing iterative training on the target vector and the post-operation predicted image for a plurality of times according to the pre-operation image and the post-operation image to obtain an image generation model of the aortic valve calcification displacement, wherein the image generation model comprises the following steps:
s741: during the first iterative training, the input of an image generation model of the aortic valve calcification displacement is a preoperative image and a target vector, wherein the target vector is obtained by original parameters according to the preoperative image;
s742: and in other iterative training except the first time, the input of the image generation model of the aortic valve calcification displacement is a postoperative predicted image obtained in the previous iterative training and a target vector obtained in the previous iterative training, wherein the target vector is determined according to a loss function obtained in the previous iterative training.
Specifically, referring to fig. 11 in combination with fig. 9, S743: a first image is obtained based on the preoperative image and the target vector, and the first image is used as a post-operative predicted image.
S744: and obtaining a loss function based on the postoperative predicted image and the postoperative image, and judging whether the loss function is smaller than a preset value.
If not, execution S745: the method includes the steps of adjusting a target vector to an update vector based on a loss function, taking the update vector as a new target vector, obtaining a second image based on the post-operation predicted image and the new target vector, and taking the second image as the post-operation predicted image.
If so, execution S746: an image generation model of the aortic valve calcification displacement is obtained.
In performing the training of steps S743 through S746, all partial parameters of the calcified region generator are frozen, and the calcified region generator is used to output an aortic valve calcified region construction loss function that is imaged after the transcatheter aortic valve replacement surgery, back-propagated to train the calcified region generator and the surgical plan encoder set. The training results in a generator and encoder set for each sampling step.
In this embodiment, the calcified region generator is a convolution generating network, and all parameters of the calcified region generator are frozen during training to ensure that the generation result of the calcified region generator is stable, so that the difference between the training result and the result of the first training (step S72 shown in fig. 9) is small.
Finally, since the semantic guidance-based diffusion model has a certain randomness in the generation process, steps S3 to S7 shown in fig. 7 are performed multiple times, and the obtained multiple predicted calcified region displacement images are superimposed to obtain a final prediction result. The results represent in the form of a heat map the possible displacement of the calcified region of the aortic valve after transcatheter aortic valve replacement under a specific surgical protocol, making the resulting predicted displacement image of the calcified region more accurate.
In other possible embodiments of the present invention, the sequence of steps of the training method for aortic valve calcification displacement image generation is not limited.
By way of example, in connection with fig. 1 to 5, an exemplary embodiment of an aortic valve calcification displacement image generation method is described below.
Referring to fig. 1, an embodiment of the present invention discloses an aortic valve calcification displacement image generation method using the training method of aortic valve calcification displacement image generation as described in the previous examples to train the generated image generation model.
Specifically, referring to fig. 1, the aortic valve calcification displacement image generation method includes:
s1: a preoperative image of the root of the valved aortic arch and the calcified region is acquired.
In particular, the preoperative image is acquired by the method as shown in fig. 5:
s01 is executed: a CTA contrast image of the aorta prior to the operation is acquired. A CTA contrast image of the aorta prior to the operation may be acquired by medical imaging methods or other possible methods.
S02: the CTA contrast image of the pre-operative aorta is segmented to obtain a pre-operative image of the valved aortic root and calcified region. The preoperative image of the valved aortic root and calcified region may be obtained by manual segmentation or algorithmic segmentation of a CTA contrast image of the preoperative aorta.
With continued reference to fig. 1, S2: and obtaining a target vector, wherein the target vector is determined according to a surgical scheme and/or patient sign parameters matched with the preoperative image to be processed.
First, referring to fig. 2, S21 is performed: the original parameters are acquired and preprocessed to obtain the first vector. Illustratively, there are two methods of obtaining the first vector,
with continued reference to fig. 2, S22 is performed: the target vector is obtained by the text encoder based on the first vector.
The target vector is used as a guide condition of the aortic valve calcification displacement image generation model, so that a target image generated on the basis of a preoperative image to be processed is more accurate and more accords with expectations.
With continued reference to fig. 1, S3 is performed: the preoperative image and the target vector are input into an image generation model of aortic valve calcification displacement to obtain a target image.
By adopting the technical scheme, the preoperative image of the aortic root with the valve and the calcified region directly and accurately generates the displacement image of the calcification of the aortic valve, namely the target image, through the image generation model of the calcification displacement of the aortic valve, so that people intuitively observe the displacement of the calcification of the aortic valve.
In other possible embodiments of the present invention, the sequence of steps of the aortic valve calcification displacement image generation method is not limited.
Fourth aspect, referring to fig. 12, the present invention provides a training system 0 for aortic valve calcification displacement image generation, comprising: an acquisition module 01 and a training module 02.
The acquisition module 01 is used for acquiring preoperative images of the root and calcified region of the aortic valve; obtaining a target vector, wherein the target vector is determined according to an operation scheme and/or patient sign parameters matched with a preoperative image;
the training module 02 is used for processing the preoperative image and the target vector through a model to be trained to obtain a post-operative predicted image; updating the target vector of the model to be trained based on the postoperative predicted image until the model training condition is met, and obtaining an image generation model of the aortic valve calcification displacement.
By adopting the technical scheme, the image predicted by the image generation training system of the aortic valve calcification displacement is accurate and is fit with reality.
In a fifth aspect, referring to fig. 13, the present invention provides an electronic device 2, comprising a memory 21, a processor 22 and a computer program stored in the memory 21 and executable on the processor 22, the processor 22 implementing the aortic valve calcification displacement image generation method in any of the embodiments described above and/or implementing the training method of any of the aortic valve calcification displacement image generation described above when executing the computer program. The memory 21 may include, for example, a system memory, a fixed nonvolatile storage medium, and the like. The system memory stores, for example, an operating system, application programs, boot Loader (Boot Loader), and other programs.
In this embodiment, the electronic device for generating the pre-operation image of the aortic valve calcification displacement image can directly and accurately generate the aortic valve calcification displacement image, so as to provide an image for intuitively observing the displacement of the aortic valve calcification for people. The image predicted by the electronic equipment is accurate and is fit with the reality.
In a sixth aspect, embodiments of the present invention further provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements any one of the aortic valve calcification displacement image generation methods described above, and/or implements any one of the training methods described above for aortic valve calcification displacement image generation.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flowchart and/or block of the flowchart illustrations and/or block diagrams, and combinations of flowcharts and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
While the invention has been shown and described with reference to certain preferred embodiments thereof, it will be understood by those skilled in the art that the foregoing is a further detailed description of the invention with reference to specific embodiments, and it is not intended to limit the practice of the invention to those descriptions. Various changes in form and detail may be made therein by those skilled in the art, including a few simple inferences or alternatives, without departing from the spirit and scope of the present invention.
Claims (18)
1. A method for generating aortic valve calcification displacement image is characterized in that,
acquiring preoperative images of the root and calcified region of the aortic valve;
obtaining a target vector, wherein the target vector is determined according to a surgical scheme and/or patient sign parameters matched with the preoperative image;
and inputting the preoperative image and the target vector into an image generation model to obtain a target image.
2. The aortic valve calcification displacement image generation method of claim 1, wherein the acquiring a target vector comprises:
acquiring and preprocessing original parameters to obtain a first vector, wherein the original parameters at least comprise the surgical scheme and/or patient sign parameters;
and obtaining a target vector through a text encoder according to the first vector.
3. The aortic valve calcification displacement image generation method of claim 2, wherein the acquiring and preprocessing of the raw parameters to obtain a first vector comprises:
obtaining the original parameters according to the preoperative image;
extracting parameters related to aortic valve calcification displacement from the original parameters as target parameters;
and constructing a first vector according to the target parameter.
4. The aortic valve calcification displacement image generation method as set forth in claim 3, wherein the extracting parameters related to the aortic valve calcification displacement from the original parameters as target parameters comprises:
extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters, and formatting the extraction parameters to obtain target parameters.
5. The aortic valve calcification displacement image generation method as set forth in claim 4, wherein the extracting a parameter related to the aortic valve calcification displacement from the original parameters as a target parameter comprises:
extracting parameters related to aortic valve calcification displacement from the original parameters as extraction parameters;
weighting the extracted parameters to obtain weighted parameters;
normalizing the weighted parameters to obtain target parameters.
6. The aortic valve calcification displacement image generation method according to any one of claims 1 to 5, wherein the image generation model of aortic valve calcification displacement adopts a diffusion model.
7. The aortic valve calcification displacement image generation method of any one of claims 1 to 5, wherein the image generation model of aortic valve calcification displacement employs conditional generation of an countermeasure network or flow model.
8. The aortic valve calcification displacement image generation method of claim 1, wherein prior to the acquiring the pre-operative image of the valved aortic root and calcified region, the method comprises:
acquiring a CTA contrast image of the aorta before operation;
and dividing the CTA contrast image of the aorta before operation to obtain a preoperation image of the root and calcified region of the aortic valve.
9. An aortic valve calcification displacement image generation system, comprising:
the acquisition module is used for acquiring preoperative images of the aortic root with the valve and the calcified region and acquiring target vectors, and the target vectors are determined according to the operation scheme and/or the patient sign parameters;
the processing module is used for inputting the preoperative image and the target vector into an image generation model so as to obtain a target image;
and the output module is used for outputting the target image.
10. A training method for aortic valve calcification displacement image generation, the training method comprising:
acquiring preoperative images of the root and calcified region of the aortic valve;
obtaining a target vector, wherein the target vector is determined according to a surgical scheme and/or patient sign parameters matched with the preoperative image;
Processing the preoperative image and the target vector through a model to be trained to obtain a post-operative predicted image;
updating the target vector of the model to be trained based on the postoperative predicted image until model training conditions are met, and obtaining an image generation model of the aortic valve calcification displacement.
11. The training method of aortic valve calcification displacement image generation of claim 10, wherein the training method further comprises:
and acquiring a postoperative image of the root and calcified region of the valved aorta corresponding to the preoperative image.
12. The training method of aortic valve calcification displacement image generation of claim 11, wherein the model training conditions include:
acquiring and preprocessing original parameters to obtain a first vector, wherein the original parameters at least comprise the surgical scheme and/or patient sign parameters;
performing calcification region image generation training on the preoperative image;
obtaining a target vector through a text encoder according to the first vector;
and performing repeated iterative training on the target vector and the postoperative predicted image according to the preoperative image and the postoperative image to obtain an image generation model of the aortic valve calcification displacement, wherein the postoperative predicted image is obtained according to the preoperative image and the target vector.
13. The training method of aortic valve calcification displacement image generation of claim 12, wherein performing multiple iterative training on the target vector, post-operative prediction image from the pre-operative image, the post-operative image to obtain an image generation model of the aortic valve calcification displacement comprises:
during the first iterative training, the input of the image generation model of the aortic valve calcification displacement is the preoperative image and the target vector, wherein the target vector is obtained by the original parameters according to the preoperative image;
and in other iterative training except the first time, the input of the image generation model of the aortic valve calcification displacement is the postoperative predicted image obtained in the previous iterative training and the target vector obtained in the previous iterative training, wherein the target vector is determined according to a loss function obtained in the previous iterative training.
14. The training method of aortic valve calcification displacement image generation of claim 13, wherein performing multiple iterative training on the target vector, post-operative prediction image from the pre-operative image, the post-operative image to obtain an image generation model of the aortic valve calcification displacement comprises:
Obtaining a first image based on the preoperative image and the target vector, and taking the first image as a post-operative predicted image;
and (3) performing iteration: obtaining a loss function based on the postoperative predicted image and the postoperative image, adjusting the target vector into an update vector based on the loss function, and taking the update vector as a new target vector;
obtaining a second image based on the post-operation predicted image and the new target vector, and taking the second image as the post-operation predicted image;
and obtaining an image generation model of the aortic valve calcification displacement until the loss function is smaller than a preset value.
15. The training method of aortic valve calcification displacement image generation of claim 11, wherein prior to the acquiring a post-operative image of the valved aortic root and calcified region corresponding to the pre-operative image, the method comprises:
acquiring a CTA contrast image of the aorta after operation;
and dividing the CTA contrast image of the aorta after operation to obtain an image of the root of the aortic valve with calcification area after operation.
16. A training system for aortic valve calcification displacement image generation, comprising:
The acquisition module is used for acquiring preoperative images of the root and calcified region of the aortic valve; obtaining a target vector, wherein the target vector is determined according to a surgical scheme and/or patient sign parameters matched with the preoperative image;
the training module is used for processing the preoperative image and the target vector through a model to be trained to obtain a post-operative predicted image; updating the target vector of the model to be trained based on the postoperative predicted image until model training conditions are met, and obtaining an image generation model of the aortic valve calcification displacement.
17. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the aortic valve calcification displacement image generation method according to any one of claims 1 to 8 and/or implements the training method of aortic valve calcification displacement image generation according to any one of claims 10 to 15 when executing the computer program.
18. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, implements the aortic valve calcification displacement image generation method according to any one of claims 1 to 8 and/or implements the training method of aortic valve calcification displacement image generation according to any one of claims 10 to 15.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310172612.8A CN116364246A (en) | 2023-02-24 | 2023-02-24 | Aortic valve calcification displacement image generation method, training method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310172612.8A CN116364246A (en) | 2023-02-24 | 2023-02-24 | Aortic valve calcification displacement image generation method, training method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116364246A true CN116364246A (en) | 2023-06-30 |
Family
ID=86911424
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310172612.8A Pending CN116364246A (en) | 2023-02-24 | 2023-02-24 | Aortic valve calcification displacement image generation method, training method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116364246A (en) |
-
2023
- 2023-02-24 CN CN202310172612.8A patent/CN116364246A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11741605B2 (en) | Method and system for image registration using an intelligent artificial agent | |
CN112997215B (en) | Method and system for automatically regionalizing blood vessels of medical images by using machine learning and image processing algorithms | |
Butoi et al. | Universeg: Universal medical image segmentation | |
van Amsterdam et al. | Multi-task recurrent neural network for surgical gesture recognition and progress prediction | |
Qin et al. | Temporal segmentation of surgical sub-tasks through deep learning with multiple data sources | |
Rezaei et al. | Whole heart and great vessel segmentation with context-aware of generative adversarial networks | |
Popescu et al. | Retinal blood vessel segmentation using pix2pix gan | |
KR102304339B1 (en) | Method and computing device for providing analysis information on intravascular ultrasound images of blood vessels using artificial neural networks | |
Wu et al. | DA-U-Net: densely connected convolutional networks and decoder with attention gate for retinal vessel segmentation | |
CN117524402A (en) | Method for analyzing endoscope image and automatically generating diagnostic report | |
Yang et al. | Efficient catheter segmentation in 3D cardiac ultrasound using slice-based FCN with deep supervision and f-score loss | |
JP7349018B2 (en) | Coronary artery segmentation method, apparatus, electronic device and computer readable storage medium | |
La Rosa | A deep learning approach to bone segmentation in CT scans | |
Fontanella et al. | Diffusion models for counterfactual generation and anomaly detection in brain images | |
CN116364246A (en) | Aortic valve calcification displacement image generation method, training method and system | |
Fakhry et al. | Elastic net regularization and gabor dictionary for classification of heart sound signals using deep learning | |
CN102697482A (en) | Conduit tracking method and system using self-adaptation to judge learning and measuring | |
CN113902689A (en) | Blood vessel center line extraction method, system, terminal and storage medium | |
CN111180046A (en) | Determining a processing sequence for processing an image | |
Li et al. | Interactive medical image segmentation with self-adaptive confidence calibration | |
Singh et al. | Normalised Gradient Gaussian Filtering for Lung Field Segmentation in X-Ray Images and Snake Segmentation | |
Zhou et al. | Deep reinforcement learning in medical imaging | |
Gullón | Retinal lesions segmentation using CNNs and adversarial training | |
CN113223104B (en) | Cardiac MR image interpolation method and system based on causal relationship | |
Guo et al. | An improved threshold method based on histogram entropy for the blood vessel segmentation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |