CN113962957A - Medical image processing method, bone image processing method, device and equipment - Google Patents

Medical image processing method, bone image processing method, device and equipment Download PDF

Info

Publication number
CN113962957A
CN113962957A CN202111228227.8A CN202111228227A CN113962957A CN 113962957 A CN113962957 A CN 113962957A CN 202111228227 A CN202111228227 A CN 202111228227A CN 113962957 A CN113962957 A CN 113962957A
Authority
CN
China
Prior art keywords
processed
feature
thermodynamic diagram
medical image
dimensional
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
Application number
CN202111228227.8A
Other languages
Chinese (zh)
Inventor
刘赫
张朗
刘鹏飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Suzhou Xiaowei Changxing Robot Co ltd
Original Assignee
Suzhou Xiaowei Changxing Robot Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Suzhou Xiaowei Changxing Robot Co ltd filed Critical Suzhou Xiaowei Changxing Robot Co ltd
Priority to CN202111228227.8A priority Critical patent/CN113962957A/en
Publication of CN113962957A publication Critical patent/CN113962957A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4007—Scaling of whole images or parts thereof, e.g. expanding or contracting based on interpolation, e.g. bilinear interpolation
    • 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
    • G06T2207/30008—Bone

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Radiology & Medical Imaging (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • Apparatus For Radiation Diagnosis (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

The application relates to a medical image processing method, a bone image processing device and bone image processing equipment. The medical image processing method includes: acquiring a to-be-processed three-dimensional medical image of a medical detection object; inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed; and determining a target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed. The bone image processing method comprises the following steps: acquiring a bone image to be processed; processing the bone image to be processed according to a medical image processing method to obtain the bone feature to be processed; and outputting the bone to-be-processed characteristics. By adopting the method, the preoperative planning work flow can be shortened, and the intelligent degree is improved.

Description

Medical image processing method, bone image processing method, device and equipment
Technical Field
The present application relates to the field of artificial intelligence technology, and in particular, to a medical image processing method, a bone image processing method, a device, and an apparatus.
Background
With the development of computer technology, doctors can be guided in the medical field by taking images of specific tissue organs or bones.
Conventionally, a physician typically selects a specific feature to be processed of a tissue organ or a bone from an acquired image, for example, the physician may determine the specific feature to be processed by drawing an auxiliary line.
However, the current physician expends energy and time to manually label the feature to be processed each time of preoperative planning, which inevitably results in long operation process and the precision of manual process labeling is difficult to maintain.
Disclosure of Invention
In view of the above, it is necessary to provide a medical image processing method, a bone image processing method, a device, and an apparatus that can shorten a preoperative planning workflow and improve an intelligent degree. A medical image processing method, the medical image processing method comprising:
acquiring a to-be-processed three-dimensional medical image of a medical detection object;
inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed;
determining a target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
In one embodiment, after acquiring the three-dimensional medical image to be processed of the medical test object, the method further includes:
performing first preprocessing on the three-dimensional medical image to be processed, wherein the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation, and is used for being input into the feature recognition model to be processed; the filtering operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to a voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is to unify dimensions of voxels in the three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
In one embodiment, the determining the target feature to be processed in the three-dimensional medical image to be processed according to the feature to be processed thermodynamic diagram includes:
filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed;
normalizing the voxels in the filtered characteristic thermodynamic diagram to be processed;
and selecting a point with a voxel value meeting the requirement in the normalized feature thermodynamic diagram to be processed as the target feature to be processed in the three-dimensional medical image to be processed.
In one embodiment, after the determining the target feature to be processed in the three-dimensional medical image according to the feature to be processed thermodynamic diagram, the method includes:
receiving an editing instruction aiming at the target feature to be processed;
and confirming the target to-be-processed characteristic according to the editing instruction, or adjusting the target to-be-processed characteristic according to the editing instruction.
A training method of a feature recognition model to be processed in the medical image processing method comprises the following steps:
acquiring a sample three-dimensional image, and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image;
obtaining an initial model, and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram;
calculating to obtain a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram;
and performing parameter iterative updating on the initial model based on the target loss function until training is completed to obtain a to-be-processed feature recognition model.
In one embodiment, the generating a real thermodynamic diagram according to the to-be-processed features of the sample labeled in the three-dimensional image of the sample comprises:
performing coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point;
calculating the distance between a point in the sample three-dimensional image and a reference point;
and generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
In one embodiment, after the acquiring the three-dimensional image of the sample, the method further includes:
performing second preprocessing on the sample three-dimensional image, wherein the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation, and is used for inputting the sample three-dimensional image into a to-be-processed feature recognition model to be trained; the image enhancement operation includes at least one of random rotation, random horizontal/vertical flipping, and random cropping.
In one embodiment, the initial model comprises a concatenation of at least one target network structure; inputting the sample three-dimensional image into the initial model to obtain a predictive thermodynamic diagram, wherein the predictive thermodynamic diagram comprises:
inputting the sample three-dimensional image into at least one target network structure of a cascade to obtain at least one predictive thermodynamic diagram;
the calculating according to the real thermodynamic diagram and the predicted thermodynamic diagram to obtain an objective loss function comprises:
calculating a target loss function of each predicted thermodynamic diagram and the real thermodynamic diagram;
the parameter iterative updating of the initial model based on the target loss function until the training is completed to obtain a to-be-processed feature recognition model comprises the following steps:
and performing parameter iterative updating on each target network structure in the initial model based on the target loss function until training is completed to obtain a to-be-processed feature recognition model.
In one embodiment, the calculating an objective loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram includes:
extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule, and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram;
calculating to obtain a first loss function according to the first voxel set and the second voxel set;
calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set;
and performing weighted calculation according to the first loss function and the second loss function to obtain a target loss function.
A bone image processing method, the bone image processing method comprising:
acquiring a bone image to be processed;
processing the bone image to be processed according to the medical image processing method or the training method to obtain the bone feature to be processed;
and outputting the bone to-be-processed characteristics.
In one embodiment, the outputting the bone to-be-processed feature comprises:
receiving an editing instruction for the bone to-be-processed feature;
and confirming the bone to-be-processed characteristic according to the editing instruction, or adjusting the bone to-be-processed characteristic according to the editing instruction.
In one embodiment, the bone treatment feature includes at least one of a lateral condyle highest point, a medial condyle lowest point, a medial condyle tangent point, and a lateral condyle tangent point.
A medical image processing apparatus, the medical image processing apparatus comprising:
the to-be-processed three-dimensional medical image acquisition module is used for acquiring a to-be-processed three-dimensional medical image of a medical detection object;
the model processing module is used for inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training to obtain a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed;
the target feature to be processed calculation module is used for determining the target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
A training device for a feature recognition model to be processed in the medical image processing device, the training device comprising:
the sample three-dimensional image acquisition module is used for acquiring a sample three-dimensional image and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image;
the model processing module is used for acquiring an initial model and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram;
the loss function calculation module is used for calculating a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram;
and the training module is used for carrying out parameter iterative updating on the initial model based on the target loss function until the training is finished to obtain a to-be-processed feature recognition model.
A bone image processing apparatus, the bone image processing apparatus comprising:
the bone image acquisition module to be processed is used for acquiring a bone image to be processed;
the bone to-be-processed feature calculation module is used for processing the bone image to be processed according to the medical image processing device to obtain bone to-be-processed features;
and the output module is used for outputting the bone to-be-processed characteristics.
A computer device comprising a memory storing a computer program and a processor implementing the steps of the method as described in any one of the above embodiments when the processor executes the computer program.
A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method as claimed in any one of the above embodiments.
According to the medical image processing method, the bone image processing device and the bone image processing equipment, the feature thermodynamic diagram to be processed is obtained by processing the three-dimensional medical image to be processed through the feature identification model to be processed, the target feature to be processed in the three-dimensional medical image to be processed is determined according to the feature thermodynamic diagram to be processed, the preoperative planning work flow is shortened, and the intelligent degree is improved.
Drawings
FIG. 1 is a diagram of an embodiment of a medical image processing method;
FIG. 2 is a flow diagram of a method of medical image processing in one embodiment;
FIG. 3 is a schematic diagram of a filtering operation in one embodiment;
FIG. 4 is a schematic diagram of a resampling operation in one embodiment;
FIG. 5 is a schematic diagram of a scale change operation in one embodiment;
FIG. 6 is a schematic flow chart of post-processing steps in one embodiment;
FIG. 7 is a flowchart illustrating a method for training a feature recognition model to be processed according to an embodiment;
FIG. 8 is a schematic diagram of classification of a sample three-dimensional image in one embodiment;
FIG. 9 is a schematic diagram of an initial model in one embodiment;
FIG. 10 is a diagram illustrating an image enhancement operation, in one embodiment;
FIG. 11 is a block diagram illustrating a target network structure, in one embodiment;
FIG. 12 is a diagram of an objective loss function calculation process in one embodiment;
FIG. 13 is a flow chart illustrating a method of medical image processing according to another embodiment;
FIG. 14 is a flow diagram illustrating a method for bone image processing according to one embodiment;
FIG. 15 is a schematic illustration of an image of a bone in one embodiment;
FIG. 16 is a block diagram showing the configuration of a medical image processing apparatus according to an embodiment;
FIG. 17 is a block diagram of a bone image processing apparatus according to an embodiment;
FIG. 18 is a diagram illustrating an internal structure of a computer device according to an embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The medical image processing method provided by the application can be applied to the application environment shown in fig. 1. Wherein the terminal 102 communicates with the medical imaging device 104 over a network. The terminal 102 may receive a to-be-processed three-dimensional medical image obtained by scanning by the medical imaging device 104, so as to input the to-be-processed three-dimensional medical image into a to-be-processed feature recognition model obtained by pre-training, and obtain a to-be-processed feature thermodynamic diagram corresponding to the to-be-processed three-dimensional medical image; and determining the target to-be-processed characteristics in the to-be-processed three-dimensional medical image according to the to-be-processed characteristic thermodynamic diagram. The preoperative planning work flow is shortened, and the intelligent degree is improved.
The terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices, and the medical imaging device 104 includes, but is not limited to, various imaging devices, such as a CT imaging device (Computed Tomography) that performs cross-sectional scans around a certain part of a human body together with a highly sensitive detector using a precisely collimated X-ray beam and reconstructs a precise three-dimensional position image of a tumor and the like through CT scanning), a magnetic resonance device (which is a Tomography device that obtains electromagnetic signals from a human body using a magnetic resonance phenomenon and reconstructs a human body information image), a Positron Emission Computed Tomography (Positron Emission Computed Tomography) device, a Positron Emission magnetic resonance imaging system (PET/MR), and the like.
In one embodiment, as shown in fig. 2, a medical image processing method is provided, which is exemplified by the application of the method to the terminal in fig. 1, and includes the following steps:
s202: acquiring a to-be-processed three-dimensional medical image of a medical detection object.
Specifically, the three-dimensional medical image to be processed is a three-dimensional image including a plurality of voxels, for example, the three-dimensional medical image to be processed may be a CTA (computed tomography angiography) volume data (for example, three-dimensional data of a human body image) image, and the terminal acquires the three-dimensional medical image to be processed scanned by the CT apparatus.
S204: inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed.
Specifically, the to-be-processed feature recognition model is a pre-trained model for recognizing the to-be-processed three-dimensional medical image to obtain a to-be-processed feature thermodynamic diagram, and the model can be formed by stacking n times based on a Vnet structure network. The probability regression analysis can be carried out on the three-dimensional medical image to be processed through the feature recognition model to be processed, so that the three-dimensional medical image to be processed generates a feature thermodynamic diagram to be processed which accords with Gaussian distribution by taking the feature to be processed as a center.
The thermodynamic diagram of the feature to be processed may be regarded as an image that maps voxels in the three-dimensional medical image to another target space, where the value of each voxel in the thermodynamic diagram is used to characterize whether the size of the feature to be processed exists at the voxel, and the size may be within a preset range, for example, 0-1, and in other embodiments, the preset range may be other values. Optionally, the characteristic thermodynamic diagram to be processed is displayed with different identifiers, such as colors, according to the change trend of the voxels.
Wherein optionally, the feature recognition model to be processed may be multi-channel, preferably different channels characterizing different types of features to be processed. Namely, the initial region corresponding to at least one feature to be processed in the three-dimensional medical image to be processed can be determined by once recognition.
S206: and determining a target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed, wherein the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
Specifically, the terminal processes the feature to be processed thermodynamic diagram to extract a target feature to be processed therein, where the terminal may acquire a position with a maximum voxel value in a target space as the target feature to be processed. In other embodiments, the terminal performs preprocessing according to the voxel value, for example, voxel value range processing, so that the voxel value in the feature thermodynamic diagram to be processed is within a certain range, and interference of an extreme value of the voxel value is avoided. In addition, the voxel values can be normalized, namely normalized to a certain range, so that the dimensions of all the voxel values are unified, and then the position with the maximum voxel value is selected as the target feature to be processed, namely the brightest point in the feature thermodynamic diagram to be processed is selected as the target feature to be processed.
The preoperative image processing includes, but is not limited to, determining a surgical reference parameter according to the identified target to-be-processed feature, for example, determining a force line of a bone according to the identified target to-be-processed feature, or determining an osteotomy range by taking the target to-be-processed feature as a reference.
According to the medical image processing method, the feature thermodynamic diagram to be processed is obtained by processing the three-dimensional medical image to be processed through the feature identification model to be processed, the target feature to be processed in the three-dimensional medical image to be processed is determined according to the feature thermodynamic diagram to be processed, the preoperative planning work flow is shortened, and the intelligent degree is improved.
In one embodiment, after acquiring the three-dimensional medical image to be processed of the medical detection object, the method further includes: performing first preprocessing on the three-dimensional medical image to be processed, wherein the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation; the data range adjustment operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to the voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is the dimension of a voxel in a three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
Specifically, in the present embodiment, the type of the first preprocessing described above is explained with reference to the drawings.
Referring to fig. 3, fig. 3 is a schematic diagram of a filtering operation in an embodiment in which voxels are filtered primarily by defining a range of voxel values of a three-dimensional medical image to be processed. Specifically, the terminal acquires a preset specific window width level, that is, a range of voxel values, and then compresses the voxel values to the range to implement filtering processing of the image. Specifically, the voxel value of each voxel in the three-dimensional medical image to be processed is compared with a specific window width level, and the voxels corresponding to the voxel values which are not in the specific window width level are deleted, so that irrelevant voxels are deleted, and the model can better extract features.
Referring to fig. 4, fig. 4 is a schematic diagram of a resampling operation in an embodiment, in which the resolutions of different data are unified through resampling, and taking fig. 4 as an example, which is illustrated in two dimensions, each black dot in fig. 4 represents a voxel, and the resampling process is implemented through interpolation, and the physical size is not changed, but the image size can be changed, so that the size of the image to be processed input to the model meets the requirement.
The normalization process is mainly to unify the values of the voxels in the image to be processed into a space in an unlimited manner, and has the function of unifying the data distribution and accelerating the network convergence.
Referring to fig. 5, fig. 5 is a schematic diagram of a scale change operation in an embodiment, where the purpose of the scale change operation is to meet the requirement of a segmentation network on an input size, and two ways of edge cropping and filling are introduced, where edge filling refers to filling a dimension of an image that does not meet the input requirement with a value of 0 so that the dimension meets the requirement. The edge cutting refers to cutting the dimension of the image which does not meet the input requirement so that the dimension meets the requirement, and the cutting mode can be that two edges of the image are cut simultaneously or only one edge is cut. The scaling is performed mainly to make the size of the image data input by the network meet the requirement of the network, and the following description will be made in conjunction with the network structure of the model.
It should be noted that, in the first preprocessing operation, the terminal may select at least one of the three-dimensional medical images to be processed for preprocessing. In a preferred embodiment, the terminal sequentially performs a filtering operation, a resampling operation, a normalizing operation and a scale changing operation on the three-dimensional medical image to be processed.
In one embodiment, determining a target feature to be processed in a three-dimensional medical image to be processed according to a feature to be processed thermodynamic diagram includes: filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed; normalizing the voxels in the filtered characteristic thermodynamic diagram to be processed; and selecting points of which the voxel values meet the requirements in the normalized feature thermodynamic diagram to be processed as target features to be processed in the three-dimensional medical image to be processed.
Specifically, referring to fig. 6, fig. 6 is a schematic flow chart of a post-processing step in an embodiment, in which after the terminal acquires the feature thermodynamic diagram to be processed, the feature thermodynamic diagram to be processed is post-processed to obtain the target feature to be processed.
The terminal firstly carries out thresholding filtering on the feature thermodynamic diagram to be processed so as to delete an extreme value in the feature thermodynamic diagram to be processed and avoid the influence of the extreme value on subsequent processing, then carries out normalization processing on the rest voxels to unify dimensions, finally obtains a feature image coordinate to be processed by finding a point index with a voxel value meeting requirements, and takes a position corresponding to the image coordinate as a target feature to be processed, wherein optionally, the terminal converts the feature image coordinate to be processed into a world coordinate after obtaining the feature image coordinate to be processed. The point where the voxel value satisfies the requirement may be a point of a maximum voxel value, or a local center position where a weighted average or a local average is maximum, and the like, which is not specifically limited herein.
In the embodiment, the thermodynamic diagram of the feature to be processed is post-processed, so that the target feature to be processed is accurately positioned.
In one embodiment, the present application further relates to a training method of a feature recognition model to be processed, and specifically, as shown in fig. 7, the training method of the feature recognition model to be processed includes:
s702: and acquiring a sample three-dimensional image, and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image.
Specifically, the sample three-dimensional image may be a pre-collected image in which the sample three-dimensional image is marked with a sample feature to be processed. The characteristic of the sample to be processed can be represented by the form of coordinates.
It should be noted that the sample three-dimensional image may be classified, for example, a part of the sample three-dimensional image is used as training data, a part of the sample three-dimensional image is used as test data, data labeling is performed on the training data, and the labeled data is divided into a training set and a verification set. As shown in fig. 8, after acquiring joint CT image data (three-dimensional), the terminal first divides the joint CT image data into training data and test data according to a specific ratio; training data needs to be manually marked on specific to-be-processed characteristic positions by doctors or medical staff with related qualifications to obtain a plurality of to-be-processed characteristic physical coordinates; then, dividing the marked training data into a training set and a verification set again according to a specific proportion, wherein each training data comprises a CT image and marked characteristic coordinates to be processed; in the network training process, the training set data is used for adjusting and optimizing model parameters; training models at different stages can be evaluated through the data of the verification set, and the model with the best performance can be obtained.
The real thermodynamic diagram is generated according to the marked sample to-be-processed features, the terminal performs coordinate conversion on position data of each CT image sample data in the sample three-dimensional image in a world coordinate system, and then the position data is mapped to the mentioned target space according to the marked sample to-be-processed features to obtain the thermodynamic diagram, which can be specifically referred to as the following.
S704: and acquiring an initial model, and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram.
Specifically, the initial model is an initial network structure of the feature recognition model to be processed, and as shown in fig. 9, fig. 9 is a schematic diagram of the initial model in one embodiment, and in this case, the initial model is composed of network stacking of Vnet structures n times.
And the terminal inputs the sample three-dimensional image into the initial model, and forward calculation is carried out to obtain a predicted thermodynamic diagram, namely the sample three-dimensional image is sequentially input into the initial model formed by stacking the Vnet structure network for n times to obtain the predicted thermodynamic diagram, wherein the form of the predicted thermodynamic diagram is similar to the characteristic thermodynamic diagram to be processed in the foregoing, and details are not repeated here.
S706: and calculating to obtain a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram.
S708: and performing parameter iterative updating on the initial model based on the target loss function until the training is completed to obtain the to-be-processed feature recognition model.
Specifically, the terminal can obtain the loss function by calculating the difference between the real thermodynamic diagram and the predicted thermodynamic diagram, that is, the degree of deviation of the features in the real thermodynamic diagram and the features in the predicted thermodynamic diagram is estimated by using the loss function. And optimizing the network through an Adam optimization algorithm according to the loss function so as to iteratively update network parameters until the loss function meets the requirement or the iteration times meet the requirement, and obtaining a to-be-processed feature recognition model.
In one embodiment, generating a real thermodynamic diagram according to the to-be-processed features of the sample labeled in the three-dimensional image of the sample comprises: performing coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point; calculating the distance between a point in the sample three-dimensional image and a reference point; and generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
The real thermodynamic diagram is generated by taking the marked characteristic coordinates of the sample to be processed as the center according to Gaussian distribution.
Specifically, the coordinates marked in the training data are usually physical coordinates of the features to be processed, and the physical coordinates of each feature to be processed are converted into image coordinates according to a mapping relationship between the features to be processed on the bone model corresponding to the training sample in advance and the marked features to be processed and a CT scanning protocol, and then a corresponding thermodynamic diagram can be generated, wherein the generation formula is as follows:
Figure BDA0003315069950000111
wherein h isgtRepresenting the generated thermodynamic diagram, piFor the characteristic coordinates, v-p, of the sample to be processediRepresenting the distance, k, of a point in the three-dimensional image of the sample from a reference pointiσ is two constants.
In one embodiment, after acquiring the three-dimensional image of the sample, the method further includes: performing second preprocessing on the sample three-dimensional image, wherein the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation; the image enhancement operation includes at least one of random rotation, random horizontal/vertical flipping, and random cropping.
Specifically, for the specific limitations of the filtering operation, the resampling operation, the normalizing operation, and the scale changing operation, reference may be made to the above, and details are not described herein again.
For the image enhancement operation, specifically, referring to fig. 10, fig. 10 is a schematic diagram of the image enhancement operation in an embodiment, in which the purpose of data enhancement is to expand the diversity of data, so as to improve the generalization and robustness of the network model. The data enhancement mainly comprises three steps: random rotation, random horizontal/vertical flipping, and random cropping. And the images are randomly rotated, turned and cut, so that sample data is expanded. Wherein the horizontal/vertical direction may be predefined by a user and is not particularly limited herein. For ease of understanding, random horizontal direction flipping means that there is half the likelihood of flipping, and there is half the likelihood of not flipping. The random vertical direction flipping is the same principle as the random horizontal direction flipping, but in the vertical direction.
It should be noted that, in the second preprocessing operation, the terminal may select at least one of the first preprocessing operation and the second preprocessing operation to preprocess the sample medical image. In a preferred embodiment, the terminal sequentially performs a filtering operation, a resampling operation, an image enhancing operation, a normalizing operation, and a scale changing operation on the sample medical image. And after the second preprocessing, inputting the sample three-dimensional image into the initial model, and adjusting parameters in the initial model through a loss function until the training is finished.
In one embodiment, the initial model includes a concatenation of at least one target network structure; inputting the sample three-dimensional image into an initial model to obtain a predictive thermodynamic diagram, wherein the predictive thermodynamic diagram comprises: inputting the sample three-dimensional image into at least one target network structure of the cascade connection to obtain at least one prediction thermodynamic diagram; calculating to obtain a target loss function according to the real thermodynamic diagram and the prediction thermodynamic diagram, wherein the target loss function comprises the following steps: calculating a target loss function of each predicted thermodynamic diagram and each real thermodynamic diagram; performing parameter iterative update on the initial model based on the target loss function until training is completed to obtain a to-be-processed feature recognition model, wherein the parameter iterative update comprises the following steps: and performing parameter iterative updating on each target network structure in the initial model based on the target loss function until training is completed to obtain the to-be-processed feature recognition model.
Specifically, referring to fig. 11, fig. 11 is a schematic structural diagram of a target network structure in an embodiment, where the target network structure is a Vnet structure, where the Vnet structure includes convolution blocks, upsampling, downsampling, convolution and layer hopping connection, where a convolution block may be composed of some different basic structures. Downsampling can be realized by using a pooling layer, and also can be realized by a convolution step size (stride 2). The upsampling may be performed by interpolation or by deconvolution. The left half of the network is subjected to down sampling to obtain image feature maps with different resolution levels, the right half of the network is subjected to up sampling by using the feature maps with different levels to recover the image, and the middle part of the network is compensated by cross-layer features. In order to realize such a connection structure, the image size must satisfy 1/2 that is just output as the input size after each down-sampling, and therefore, the input image needs to be adaptively adjusted, that is, the above scale change operation.
In the embodiment, in order to accelerate the convergence of network training, the predicted thermodynamic diagram output by each target network structure and the real thermodynamic diagram calculate a loss function to play a role of deep supervision.
In one embodiment, the calculating the target loss function according to the real thermodynamic diagram and the predictive thermodynamic diagram includes: extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule, and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram; calculating to obtain a first loss function according to the first voxel set and the second voxel set; calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set; and performing weighted calculation according to the first loss function and the second loss function to obtain a target loss function.
Specifically, referring to fig. 12, fig. 12 is a schematic diagram of the calculation of the target loss function in an embodiment, in which a calculation manner of black and white voxel separation is provided, that is, firstly, for a real thermodynamic diagram generated according to the feature to be processed of the sample, a set of voxels greater than 0 (white) is extracted as a first positive set of voxels, and voxels less than or equal to 0 (black) are extracted as a first negative set of voxels; then extracting a second positive voxel set and a second negative voxel set on the prediction thermodynamic diagram by utilizing the black and white voxel range of the real thermodynamic diagram; the computational loss between the first set of voxels and the second set of voxels is lposThe computational loss between the first negative voxel set and the second negative voxel set is lnegThen, carrying out weighted summation to obtain final loss; wherein wpos，wnegFor the weights, dynamic adjustments may be made during the training process.
In the network training process, the loss function of the network output and the real value is directly calculated, and then the network parameters are updated by using an optimization algorithm. After the network training is completed, in the using process, the output value of the network forward calculation needs to be post-processed to obtain the expected result, that is, the coordinates of the feature to be processed, and the specific post-processing mode can be referred to above.
In order to make those skilled in the art fully understand the present application, the following describes the medical image processing method of the present application in detail with reference to fig. 13, where the terminal performs data collection to obtain a sample three-dimensional image, preprocesses the sample three-dimensional image according to a second preprocessing, inputs the preprocessed sample three-dimensional image into an initial model, performs forward calculation on network parameters to obtain a predicted thermodynamic diagram, and if the predicted thermodynamic diagram is in a training process, calculates a target loss function according to a real thermodynamic diagram, updates the network parameters based on the target loss function, and if the predicted thermodynamic diagram is not in the training process, performs post-processing to obtain feature coordinates to be processed.
In one embodiment, as shown in fig. 14, a bone image processing method is provided, which is exemplified by the method applied to the terminal in fig. 1, and includes the following steps:
s1402: and acquiring an image of the bone to be processed.
In particular, the bone image to be processed is a joint CT three-dimensional image, preferably a knee joint. In other embodiments, the bone image to be processed may be other bones, and is not limited in particular.
S1404: according to the medical image processing method of any one of the embodiments, the bone image to be processed is processed to obtain the bone feature to be processed.
Specifically, the calculation of the bone feature to be processed may refer to the above calculation method of the target feature to be processed, which is not described herein again.
S1406: and outputting the bone to-be-processed characteristics.
Specifically, after the bone to-be-processed features are obtained through calculation, the terminal may mark the corresponding bone to-be-processed features in the to-be-processed bone image, where the standard modes include, but are not limited to, color, icon, graph, and the like.
Wherein, taking knee joint femoral condyle anterior-posterior osteotomy as an example, doctors need to know the medial-lateral posterior condylar line, the through condylar line and the anterior-posterior axis; therefore, the positions of some features to be processed need to be accurately positioned, so that the features to be processed of the bone include, but are not limited to, lateral condyle highest points, medial condyle lowest points, medial condyle tangent points, lateral condyle tangent points and the like, so that the preoperative plan can draw relevant joint lines according to the features to be processed to prepare for bone cutting; FIG. 15 is a schematic view of the femoral condyle access line, wherein the condyle access line is formed by connecting the lateral condyle maximum and the medial condyle minimum. In other embodiments, the joint line may be generated directly from the feature to be processed, without manual drawing by the physician.
In the embodiment, the feature thermodynamic diagram to be processed is obtained by processing the bone image to be processed through the feature identification model to be processed, and the feature to be processed of the bone in the bone image to be processed is determined according to the feature thermodynamic diagram to be processed, so that the preoperative planning work flow is shortened, and the intelligent degree is improved.
In one embodiment, after outputting the bone to-be-processed feature, the method includes: receiving an editing instruction aiming at the bone to-be-processed characteristic; and confirming the bone to-be-processed characteristics according to the editing instruction, or adjusting the bone to-be-processed characteristics according to the editing instruction.
Optionally, in this embodiment, after the bone to-be-processed features are calculated by using a deep learning algorithm, the doctor may perform two-degree confirmation, and even if a slight deviation occurs, only simple fine adjustment is required, so that the workload of preoperative planning of the doctor is greatly reduced.
It should be understood that, although the steps in the flowcharts of fig. 2, 6, 7, 13, and 14 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2, 6, 7, 13, and 14 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed alternately or alternately with other steps or at least some of the other steps.
In one embodiment, as shown in fig. 16, there is provided a medical image processing apparatus including: a to-be-processed three-dimensional medical image acquisition module 1601, a model processing module 1602, and a target to-be-processed feature calculation module 1603, wherein:
a to-be-processed three-dimensional medical image acquisition module 1601 configured to acquire a to-be-processed three-dimensional medical image of a medical detection object;
the model processing module 1602 is configured to input the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, so as to obtain a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed;
a target feature to be processed calculation module 1603, configured to determine a target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
In one embodiment, the medical image processing apparatus further includes:
the first preprocessing module is used for performing first preprocessing on the three-dimensional medical image to be processed, and the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation; the data range adjustment operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to the voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is the dimension of a voxel in a three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
In one embodiment, the target pending feature calculation module 1603 may include:
the filtering unit is used for filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed;
the normalization unit is used for performing normalization processing on voxels in the filtered characteristic thermodynamic diagram to be processed;
and the selecting unit is used for selecting points with voxel values meeting requirements in the normalized feature thermodynamic diagram to be processed as target features to be processed in the three-dimensional medical image to be processed.
In one embodiment, the apparatus further includes:
the first receiving module is used for receiving an editing instruction aiming at the target to-be-processed characteristic;
and the first adjusting module is used for confirming the target to-be-processed characteristic according to the editing instruction or adjusting the target to-be-processed characteristic according to the editing instruction.
In one embodiment, the medical image processing apparatus further includes:
the sample three-dimensional image acquisition module is used for acquiring a sample three-dimensional image and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image;
the prediction module is used for acquiring an initial model and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram;
the loss function calculation module is used for calculating a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram;
and the training module is used for carrying out parameter iterative updating on the initial model based on the target loss function until the training is finished to obtain the to-be-processed feature recognition model.
In one embodiment, the sample three-dimensional image obtaining module includes:
the coordinate conversion unit is used for carrying out coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point;
a distance calculation unit for calculating a distance between a point in the sample three-dimensional image and a reference point;
and the generating unit is used for generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
In one embodiment, the medical image processing apparatus further includes:
the second preprocessing module is used for performing second preprocessing on the sample three-dimensional image, and the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation; the image enhancement operation includes at least one of random rotation, random horizontal flipping, and random cropping.
In one embodiment the initial model comprises a concatenation of at least one target network structure; the prediction module is further used for inputting the sample three-dimensional image into at least one target network structure in cascade connection to obtain at least one prediction thermodynamic diagram;
the loss function calculation module is further used for calculating a target loss function of each of the predicted thermodynamic diagrams and the real thermodynamic diagrams;
the training module is further used for carrying out parameter iterative updating on each target network structure in the initial model based on the target loss function until the training is completed to obtain the to-be-processed feature recognition model.
In one embodiment, the loss function calculating module includes:
the voxel extraction unit is used for extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram;
the first calculation unit is used for calculating to obtain a first loss function according to the first voxel set and the second voxel set;
the second calculation unit is used for calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set;
and the third calculating unit is used for carrying out weighting calculation according to the first loss function and the second loss function to obtain a target loss function.
In one embodiment, as shown in fig. 17, there is provided a bone image processing apparatus including: a to-be-processed bone image acquisition module 1701, a bone to-be-processed feature calculation module 1702 and an output module 1703, wherein:
a to-be-processed bone image acquisition module 1701 for acquiring a to-be-processed bone image;
a bone feature to be processed calculation module 1702, configured to process the bone image to be processed according to the medical image processing apparatus in any of the above embodiments to obtain a bone feature to be processed;
and an output module 1703, configured to output the bone to-be-processed feature.
In one embodiment, the bone image processing apparatus further includes:
the second receiving module is used for receiving an editing instruction aiming at the bone to-be-processed characteristic;
and the second adjusting module is used for confirming the bone to-be-processed characteristics according to the editing instruction or adjusting the bone to-be-processed characteristics according to the editing instruction.
In one embodiment, the bone treatment feature includes at least one of a lateral condyle maximum, a medial condyle minimum, a medial condyle tangent point, and a lateral condyle tangent point.
For specific limitations of the medical image processing apparatus and the bone image processing apparatus, reference may be made to the above limitations of the medical image processing method and the bone image processing method, which are not described herein again. The modules in the medical image processing device and the bone image processing device can be wholly or partially realized by software, hardware and a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.
In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as shown in fig. 18. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The communication interface of the computer device is used for carrying out wired or wireless communication with an external terminal, and the wireless communication can be realized through WIFI, an operator network, NFC (near field communication) or other technologies. The computer program is executed by a processor to implement a medical image processing method, a bone image processing method. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 18 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program: acquiring a to-be-processed three-dimensional medical image of a medical detection object; inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed; determining target to-be-processed characteristics in the to-be-processed three-dimensional medical image according to the to-be-processed characteristic thermodynamic diagram; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
In one embodiment, the processor, when executing the computer program, further comprises, after acquiring a to-be-processed three-dimensional medical image of a medical examination object, the method further comprising: performing first preprocessing on the three-dimensional medical image to be processed, wherein the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation; the data range adjustment operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to the voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is the dimension of a voxel in a three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
In one embodiment, the determination of the target feature to be processed in the three-dimensional medical image to be processed according to the feature to be processed thermodynamic diagram, which is realized when the processor executes the computer program, comprises: filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed; normalizing the voxels in the filtered characteristic thermodynamic diagram to be processed; and selecting points of which the voxel values meet the requirements in the normalized feature thermodynamic diagram to be processed as target features to be processed in the three-dimensional medical image to be processed.
In one embodiment, the processor, when executing the computer program, further performs the steps of: receiving an editing instruction aiming at the target feature to be processed; and confirming the target to-be-processed characteristic according to the editing instruction, or adjusting the target to-be-processed characteristic according to the editing instruction.
In one embodiment, the processor, when executing the computer program, further performs the steps of: acquiring a sample three-dimensional image, and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image; obtaining an initial model, and inputting a sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram; calculating according to the real thermodynamic diagram and the predicted thermodynamic diagram to obtain a target loss function; and performing parameter iterative updating on the initial model based on the target loss function until the training is completed to obtain the to-be-processed feature recognition model.
In one embodiment, the generating of the real thermodynamic diagram according to the to-be-processed features of the sample marked in the three-dimensional image of the sample, which is realized when the processor executes the computer program, comprises: performing coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point; calculating the distance between a point in the sample three-dimensional image and a reference point; and generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
In one embodiment, the processor, when executing the computer program, further comprises, after acquiring the three-dimensional image of the sample: performing second preprocessing on the sample three-dimensional image, wherein the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation; the image enhancement operation includes at least one of random rotation, random horizontal flipping, and random cropping.
In one embodiment, the initial model implemented when the computer program is executed by the processor comprises a concatenation of at least one target network structure; inputting the sample three-dimensional image into the initial model to obtain a predictive thermodynamic diagram, which is realized when the processor executes the computer program, comprises the following steps: inputting the sample three-dimensional image into at least one target network structure of the cascade connection to obtain at least one prediction thermodynamic diagram; calculating a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram, which is realized when the processor executes the computer program, and comprises the following steps: calculating a target loss function of each predicted thermodynamic diagram and each real thermodynamic diagram; the method for performing parameter iterative update on the initial model based on the target loss function when the processor executes the computer program until training is completed to obtain the to-be-processed feature recognition model comprises the following steps: and performing parameter iterative updating on each target network structure in the initial model based on the target loss function until training is completed to obtain the to-be-processed feature recognition model.
In one embodiment, the calculation of the target loss function from the true thermodynamic diagram and the predicted thermodynamic diagram, implemented when the processor executes the computer program, includes: extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule, and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram; calculating to obtain a first loss function according to the first voxel set and the second voxel set; calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set; and performing weighted calculation according to the first loss function and the second loss function to obtain a target loss function.
In one embodiment, a computer device is provided, comprising a memory and a processor, the memory having a computer program stored therein, the processor implementing the following steps when executing the computer program: acquiring a bone image to be processed; processing a bone image to be processed according to the medical image processing method in any one of the embodiments to obtain a bone feature to be processed; and outputting the bone to-be-processed characteristics.
In one embodiment, the outputting of the bone to be processed feature implemented when the processor executes the computer program comprises: receiving an editing instruction aiming at the bone to-be-processed characteristic; and confirming the bone to-be-processed characteristics according to the editing instruction, or adjusting the bone to-be-processed characteristics according to the editing instruction.
In one embodiment, the bone preparation features involved in the execution of the computer program by the processor include at least one of a lateral condyle maximum, a medial condyle minimum, a medial condyle tangent point, and a lateral condyle tangent point.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of: acquiring a to-be-processed three-dimensional medical image of a medical detection object; inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed; determining target to-be-processed characteristics in the to-be-processed three-dimensional medical image according to the to-be-processed characteristic thermodynamic diagram; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
In one embodiment, the computer program, when executed by a processor, further comprises, after acquiring a three-dimensional medical image to be processed of a medical examination object: performing first preprocessing on the three-dimensional medical image to be processed, wherein the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation; the data range adjustment operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to the voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is the dimension of a voxel in a three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
In one embodiment, the determination of a target feature to be processed in a three-dimensional medical image to be processed from a feature to be processed thermodynamic diagram, implemented when a computer program is executed by a processor, comprises: filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed; normalizing the voxels in the filtered characteristic thermodynamic diagram to be processed; and selecting points of which the voxel values meet the requirements in the normalized feature thermodynamic diagram to be processed as target features to be processed in the three-dimensional medical image to be processed.
In one embodiment, the computer program when executed by the processor further performs the steps of: receiving an editing instruction aiming at the target feature to be processed; and confirming the target to-be-processed characteristic according to the editing instruction, or adjusting the target to-be-processed characteristic according to the editing instruction.
In one embodiment, the computer program when executed by the processor further performs the steps of: acquiring a sample three-dimensional image, and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image; obtaining an initial model, and inputting a sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram; calculating according to the real thermodynamic diagram and the predicted thermodynamic diagram to obtain a target loss function; and performing parameter iterative updating on the initial model based on the target loss function until the training is completed to obtain the to-be-processed feature recognition model.
In one embodiment, the generation of a real thermodynamic diagram from a sample feature to be processed labeled in a three-dimensional image of a sample, implemented when a computer program is executed by a processor, includes: performing coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point; calculating the distance between a point in the sample three-dimensional image and a reference point; and generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
In one embodiment, the computer program, when executed by the processor, further comprises, after acquiring a three-dimensional image of the sample: performing second preprocessing on the sample three-dimensional image, wherein the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation; the image enhancement operation includes at least one of random rotation, random horizontal flipping, and random cropping.
In one embodiment, an initial model implemented by a computer program when executed by a processor includes a concatenation of at least one target network structure; inputting a three-dimensional image of a sample into an initial model to obtain a predictive thermodynamic diagram, the computer program being implemented when executed by a processor, comprising: inputting the sample three-dimensional image into at least one target network structure of the cascade connection to obtain at least one prediction thermodynamic diagram; the calculation of the target loss function from the true thermodynamic diagram and the predicted thermodynamic diagram, which is performed when the computer program is executed by the processor, includes: calculating a target loss function of each predicted thermodynamic diagram and each real thermodynamic diagram; when the computer program is executed by the processor, the parameter iterative updating is carried out on the initial model based on the target loss function until the training is completed to obtain the feature recognition model to be processed, and the method comprises the following steps: and performing parameter iterative updating on each target network structure in the initial model based on the target loss function until training is completed to obtain the to-be-processed feature recognition model.
In one embodiment, the calculation of the objective loss function from the true thermodynamic diagram and the predicted thermodynamic diagram, when performed by the processor, comprises: extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule, and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram; calculating to obtain a first loss function according to the first voxel set and the second voxel set; calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set; and performing weighted calculation according to the first loss function and the second loss function to obtain a target loss function.
In one embodiment, a computer-readable storage medium is provided, having a computer program stored thereon, which when executed by a processor, performs the steps of: acquiring a bone image to be processed; processing a bone image to be processed according to the medical image processing method in any one of the embodiments to obtain a bone feature to be processed; and outputting the bone to-be-processed characteristics.
In one embodiment, the computer program, when executed by a processor, implements outputting the bone to-be-processed feature, comprising: receiving an editing instruction aiming at the bone to-be-processed characteristic; and confirming the bone to-be-processed characteristics according to the editing instruction, or adjusting the bone to-be-processed characteristics according to the editing instruction.
In one embodiment, the computer program when executed by the processor is directed to a bone preparation feature including at least one of a lateral condyle apex, a medial condyle tangent point, and a lateral condyle tangent point.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (17)

1. A medical image processing method, characterized in that the medical image processing method comprises:
acquiring a to-be-processed three-dimensional medical image of a medical detection object;
inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training, and obtaining a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed;
determining a target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
2. The medical image processing method according to claim 1, further comprising, after the acquiring the three-dimensional medical image to be processed of the medical test object:
performing first preprocessing on the three-dimensional medical image to be processed, wherein the first preprocessing comprises at least one of filtering operation, resampling operation, normalization operation and scale change operation, and is used for being input into the feature recognition model to be processed; the filtering operation is to perform voxel filtering operation on the three-dimensional medical image to be processed according to a voxel value range; the resampling operation is to perform interpolation operation on voxels in the three-dimensional medical image to be processed; the normalization operation is to unify dimensions of voxels in the three-dimensional medical image to be processed; the scale change operation is to adjust an image size of the three-dimensional medical image to be processed.
3. The medical image processing method according to claim 1, wherein the determining the target feature to be processed in the three-dimensional medical image to be processed according to the feature to be processed thermodynamic diagram comprises:
filtering the voxels in the characteristic thermodynamic diagram to be processed according to the voxel values in the characteristic thermodynamic diagram to be processed;
normalizing the voxels in the filtered characteristic thermodynamic diagram to be processed;
and selecting a point with a voxel value meeting the requirement in the normalized feature thermodynamic diagram to be processed as the target feature to be processed in the three-dimensional medical image to be processed.
4. A medical image processing method according to any one of claims 1 to 3, wherein after determining the target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed, the method comprises:
receiving an editing instruction aiming at the target feature to be processed;
and confirming the target to-be-processed characteristic according to the editing instruction, or adjusting the target to-be-processed characteristic according to the editing instruction.
5. A training method for a feature recognition model to be processed in the medical image processing method of any one of claims 1 to 4, wherein the training method comprises:
acquiring a sample three-dimensional image, and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image;
obtaining an initial model, and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram;
calculating to obtain a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram;
and performing parameter iterative updating on the initial model based on the target loss function until training is completed to obtain a to-be-processed feature recognition model.
6. The training method of claim 5, wherein the generating a real thermodynamic diagram according to the to-be-processed features of the sample labeled in the three-dimensional image of the sample comprises:
performing coordinate conversion on a point corresponding to the marked sample to-be-processed feature in the sample three-dimensional image to obtain a reference point;
calculating the distance between a point in the sample three-dimensional image and a reference point;
and generating a real thermodynamic diagram according to the distance and the Gaussian distribution.
7. The training method of claim 5, wherein after the obtaining the three-dimensional image of the sample, further comprising:
performing second preprocessing on the sample three-dimensional image, wherein the second preprocessing comprises at least one of filtering operation, image enhancement operation, resampling operation, normalization operation and scale change operation, and is used for inputting the sample three-dimensional image into a to-be-processed feature recognition model to be trained; the image enhancement operation includes at least one of random rotation, random horizontal/vertical flipping, and random cropping.
8. The training method of claim 5, wherein the initial model comprises a concatenation of at least one target network structure; inputting the sample three-dimensional image into the initial model to obtain a predictive thermodynamic diagram, wherein the predictive thermodynamic diagram comprises:
inputting the sample three-dimensional image into at least one target network structure of a cascade to obtain at least one predictive thermodynamic diagram;
the calculating according to the real thermodynamic diagram and the predicted thermodynamic diagram to obtain an objective loss function comprises:
calculating a target loss function of each predicted thermodynamic diagram and the real thermodynamic diagram;
the parameter iterative updating of the initial model based on the target loss function until the training is completed to obtain a to-be-processed feature recognition model comprises the following steps:
and performing parameter iterative updating on each target network structure in the initial model based on the target loss function until training is completed to obtain a to-be-processed feature recognition model.
9. The training method of claim 5, wherein the calculating an objective loss function from the real thermodynamic diagram and the predictive thermodynamic diagram comprises:
extracting a first positive voxel set and a first negative voxel set in the real thermodynamic diagram according to a preset rule, and extracting a second positive voxel set and a second negative voxel set from the predictive thermodynamic diagram;
calculating to obtain a first loss function according to the first voxel set and the second voxel set;
calculating to obtain a second loss function according to the first negative voxel set and the second negative voxel set;
and performing weighted calculation according to the first loss function and the second loss function to obtain a target loss function.
10. A bone image processing method, characterized by comprising:
acquiring a bone image to be processed;
the medical image processing method according to any one of claims 1 to 4 or the training method according to any one of claims 5 to 9, wherein the bone image to be processed is processed to obtain a bone feature to be processed;
and outputting the bone to-be-processed characteristics.
11. A bone image processing method according to claim 10, wherein said outputting said bone feature to be processed comprises:
receiving an editing instruction for the bone to-be-processed feature;
and confirming the bone to-be-processed characteristic according to the editing instruction, or adjusting the bone to-be-processed characteristic according to the editing instruction.
12. A bone image processing method as recited in claim 10, wherein the bone feature to be processed includes at least one of a lateral condyle highest point, a medial condyle lowest point, a medial condyle tangent point, and a lateral condyle tangent point.
13. A medical image processing apparatus, characterized in that the medical image processing apparatus comprises:
the to-be-processed three-dimensional medical image acquisition module is used for acquiring a to-be-processed three-dimensional medical image of a medical detection object;
the model processing module is used for inputting the three-dimensional medical image to be processed into a feature recognition model to be processed obtained through pre-training to obtain a feature thermodynamic diagram to be processed corresponding to the three-dimensional medical image to be processed;
the target feature to be processed calculation module is used for determining the target feature to be processed in the three-dimensional medical image to be processed according to the feature thermodynamic diagram to be processed; the target feature to be processed is located in a three-dimensional image space corresponding to the medical detection object, so that preoperative image processing is performed on the three-dimensional medical image to be processed based on the target feature to be processed.
14. A training apparatus for a feature recognition model to be processed in the medical image processing apparatus as set forth in claim 13, wherein the training apparatus comprises:
the sample three-dimensional image acquisition module is used for acquiring a sample three-dimensional image and generating a real thermodynamic diagram according to the to-be-processed characteristics of the sample marked in the sample three-dimensional image;
the model processing module is used for acquiring an initial model and inputting the sample three-dimensional image into the initial model to obtain a prediction thermodynamic diagram;
the loss function calculation module is used for calculating a target loss function according to the real thermodynamic diagram and the predicted thermodynamic diagram;
and the training module is used for carrying out parameter iterative updating on the initial model based on the target loss function until the training is finished to obtain a to-be-processed feature recognition model.
15. A bone image processing apparatus characterized by comprising:
the bone image acquisition module to be processed is used for acquiring a bone image to be processed;
a bone feature to be processed calculation module, configured to process the bone image to be processed according to the medical image processing apparatus of claim 13 to obtain a bone feature to be processed;
and the output module is used for outputting the bone to-be-processed characteristics.
16. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor, when executing the computer program, implements the steps of the method of any of claims 1 to 4 or 5 to 9 or 10 to 12.
17. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4 or 5 to 9 or 10 to 12.
CN202111228227.8A 2021-10-21 2021-10-21 Medical image processing method, bone image processing method, device and equipment Pending CN113962957A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111228227.8A CN113962957A (en) 2021-10-21 2021-10-21 Medical image processing method, bone image processing method, device and equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111228227.8A CN113962957A (en) 2021-10-21 2021-10-21 Medical image processing method, bone image processing method, device and equipment

Publications (1)

Publication Number Publication Date
CN113962957A true CN113962957A (en) 2022-01-21

Family

ID=79465572

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111228227.8A Pending CN113962957A (en) 2021-10-21 2021-10-21 Medical image processing method, bone image processing method, device and equipment

Country Status (1)

Country Link
CN (1) CN113962957A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115035145A (en) * 2022-05-05 2022-09-09 深圳市铱硙医疗科技有限公司 Blood vessel and bone segmentation method and device, computer device and storage medium

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115035145A (en) * 2022-05-05 2022-09-09 深圳市铱硙医疗科技有限公司 Blood vessel and bone segmentation method and device, computer device and storage medium

Similar Documents

Publication Publication Date Title
US11593943B2 (en) RECIST assessment of tumour progression
CN109690554B (en) Method and system for artificial intelligence based medical image segmentation
CN106682435B (en) System and method for automatically detecting lesion in medical image through multi-model fusion
RU2677764C2 (en) Registration of medical images
JP2021530061A (en) Image processing methods and their devices, electronic devices and computer-readable storage media
CN109754396B (en) Image registration method and device, computer equipment and storage medium
CN111862249A (en) System and method for generating canonical imaging data for medical image processing using deep learning
CN109124662B (en) Rib center line detection device and method
CN107220965B (en) Image segmentation method and system
KR102228087B1 (en) Method and apparatus for segmentation of specific cartilage in medical image
JP2006006359A (en) Image generator, image generator method, and its program
JP2006325937A (en) Image determination device, image determination method, and program therefor
CN107424162B (en) Image segmentation method and system
CN107220984B (en) Image segmentation method, image segmentation system and image segmentation device
CN115861656A (en) Method, apparatus and system for automatically processing medical images to output an alert
Saad et al. A survey on training challenges in generative adversarial networks for biomedical image analysis
CN112950684B (en) Target feature extraction method, device, equipment and medium based on surface registration
CN112200780B (en) Bone tissue positioning method, device, computer equipment and storage medium
CN113962957A (en) Medical image processing method, bone image processing method, device and equipment
CN107230211B (en) Image segmentation method and system
CN113160199A (en) Image recognition method and device, computer equipment and storage medium
CN111918611B (en) Method for controlling abnormal display of chest X-ray image, recording medium and apparatus
JP2017189394A (en) Information processing apparatus and information processing system
CN114881930A (en) 3D target detection method, device, equipment and storage medium based on dimension reduction positioning
CN115546089A (en) Medical image segmentation method, pathological image processing method, device and equipment

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