CN111369463B - Head low-dose CT image calcification point retaining method based on deep learning - Google Patents
Head low-dose CT image calcification point retaining method based on deep learning Download PDFInfo
- Publication number
- CN111369463B CN111369463B CN202010142001.5A CN202010142001A CN111369463B CN 111369463 B CN111369463 B CN 111369463B CN 202010142001 A CN202010142001 A CN 202010142001A CN 111369463 B CN111369463 B CN 111369463B
- Authority
- CN
- China
- Prior art keywords
- image
- dose
- head
- network
- low
- 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.)
- Active
Links
- 230000002308 calcification Effects 0.000 title claims abstract description 88
- 238000000034 method Methods 0.000 title claims abstract description 35
- 238000013135 deep learning Methods 0.000 title claims abstract description 22
- 208000004434 Calcinosis Diseases 0.000 claims abstract description 87
- 230000009467 reduction Effects 0.000 claims abstract description 35
- 238000011084 recovery Methods 0.000 claims abstract description 9
- 238000012549 training Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 9
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 4
- 238000003745 diagnosis Methods 0.000 abstract description 7
- 238000012545 processing Methods 0.000 abstract description 3
- 230000000717 retained effect Effects 0.000 abstract 2
- 230000014759 maintenance of location Effects 0.000 abstract 1
- 238000002591 computed tomography Methods 0.000 description 200
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 238000004321 preservation Methods 0.000 description 4
- 230000005855 radiation Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 206010028980 Neoplasm Diseases 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 201000011510 cancer Diseases 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000007429 general method Methods 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 231100000915 pathological change Toxicity 0.000 description 1
- 230000036285 pathological change Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/50—Image enhancement or restoration using two or more images, e.g. averaging or subtraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10072—Tomographic images
- G06T2207/10081—Computed x-ray tomography [CT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20212—Image combination
- G06T2207/20221—Image fusion; Image merging
-
- 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
- G06T2207/30204—Marker
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Apparatus For Radiation Diagnosis (AREA)
Abstract
The invention provides a method for reserving head low-dose CT image calcifications based on deep learning, relates to the technical field of CT image processing, and mainly comprises head low-dose CT image denoising and head CT image calcifications recovery after denoising. After the noise reduction of the head low-dose CT image, although the noise level of the image is reduced, calcification points which are similar to the noise points are weakened or eliminated. Since head calcification is one of the important criteria for diagnosis by physicians, it is highly desirable to preserve calcification while reducing the noise level of low-dose CT images of the head. The deep learning method is applied to the retention of the calcifications in the low-dose head CT image, so that the noise and the artifacts in the low-dose head CT image can be removed, the calcifications which are very similar to the noise can be retained, the quality of the head CT image is improved, the key features in the image are retained, and the diagnosis by a doctor is facilitated.
Description
Technical Field
The invention relates to a head low-dose CT image calcification point retaining method based on deep learning, and belongs to the technical field of CT image processing.
Background
Computed Tomography (CT) utilizes a precisely collimated X-ray beam and a highly sensitive detector to perform cross-sectional scanning one by one around a certain part of a human body, has the characteristics of fast scanning time, clear images and the like, can be used for the examination of various diseases, and has become one of the most important examination tools in hospitals nowadays.
With the increasing use of CT and its disadvantages, some studies in recent years have shown that the radiation generated by X-ray scanning may adversely affect the health of patients, and if the scanning dose is not properly controlled, it may induce pathological changes such as cancer. There are millions of CT exams worldwide each year, and the potential hazard to patients from the radiation produced by CT scans is of increasing concern. Therefore, the development of new techniques capable of reducing the radiation dose to the patient in the CT scan detection has become an urgent need in the medical field.
However, reducing the radiation dose in CT scan testing tends to increase image noise and artifacts, thereby affecting physician interpretation and diagnosis. Therefore, designing better low-dose CT image reconstruction or image processing methods has become a problem of great concern. With the rapid development of deep learning technology in recent years, a new idea is provided for low-dose CT image reconstruction. At present, a great deal of research is carried out on applying deep learning methods such as a Convolutional Neural Network (CNN) and a Generative Adaptive Network (GAN) to low-dose CT image reconstruction, and the methods can effectively remove image artifacts and noise and can better retain key information of an image. The principle of the deep learning method is mainly to minimize mean-squared-error between the low-dose CT image and the high-dose CT image, so that the signal-to-noise ratio (SNR) of the reconstructed low-dose CT image and the SNR of the reconstructed high-dose CT image reach the same level, and the definition of the low-dose CT image is improved.
However, there is a structure, i.e., calcifications, in the CT image of the head that is very similar to the noise point characteristics (the size of the CT value and the number of occupied pixels). When the deep learning network is used for denoising the head low-dose image, the network can wrongly treat the calcifications as noise points because the network cannot distinguish the noise points from the calcifications, and after denoising the head low-dose CT image with the calcifications, the calcifications information is weakened or even removed. If the original tissue in the image is lost in order to improve the quality of the head low-dose CT image, the purpose of denoising the head low-dose CT image is violated (the quality of the image is improved by removing some unnecessary noise and artifacts, and the diagnosis of a doctor is facilitated). Head calcification is one of the important bases for diagnosing whether a patient has a lesion, and if the information of the calcification is weakened or removed to cause the result of missed diagnosis, the significance of low-dose CT image reconstruction is lost. It is therefore important to preserve the original calcification in the image while de-noising the low dose CT image of the head.
In other words, in the prior art, identification and preservation of calcifications are mostly performed on a high-dose CT image, and noise points with similar characteristics to the calcifications generally do not exist on the high-dose CT image, so that identification and preservation of calcifications are convenient. The general method comprises the steps of firstly calculating the CT value of each pixel point on a CT image, and judging whether each pixel point is a calcification pixel point according to a set threshold value; and then, connecting all the calcification pixel points in the image into blocks by using a connecting component algorithm to obtain calcification points, and finally, marking and retaining the obtained calcification points. However, the method is only suitable for the CT image with normal dose, and cannot be applied to the marking and the preservation of the calcification points in the low-dose CT image. It is also mentioned that there are many noise points in the low-dose CT image of the head, and the characteristics (the number of pixels and the CT value) of these noise points are very similar to those of the calcifications, and the calcifications and the noise points cannot be accurately distinguished only by the above method.
The present application was made based on this.
Disclosure of Invention
In order to solve the above-mentioned defects in the prior art, the invention provides a method for preserving calcifications in a low-dose CT image of a head based on deep learning, which can not only remove noise and artifacts from the low-dose CT image of the head, but also preserve calcifications very similar to noise points, that is, preserve key features in the image while improving the quality of the low-dose CT image of the head.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
the invention applies the deep learning method to the preservation of the calcification of the head low-dose CT image, and preserves the calcification in the original head low-dose CT image while reducing the noise of the head low-dose CT image. The method mainly comprises two parts, wherein one part is used for denoising the head low-dose CT image, and the other part is used for restoring the calcifications of the head CT image after denoising. Wherein, the low-dose CT image noise reduction specifically comprises the following steps: obtaining an original head low-dose CT image and an original head high-dose CT image corresponding to the original head low-dose CT image; data preprocessing: dividing the head low-dose CT image and the head high-dose CT image into image blocks, and normalizing the CT values of the low-dose CT image block and the high-dose CT image block obtained by division; constructing a first network model and initializing network parameters; training a network model I: inputting the head low-dose CT image block and the head high-dose CT image block into a first network model to train a network, and storing the optimal parameters of the first network model when the loss function value of the first network model reaches a set threshold value; denoising low-dose CT images: setting the parameters of the network model I as the stored optimal parameters by using the optimal parameters of the network model I, inputting an original head low-dose CT image into the network model I, and outputting the original head low-dose CT image after noise reduction by a network;
after the noise of the head low-dose CT image is reduced, not only the noise in the image is removed, but also calcification points which are similar to the noise points are removed. Because head calcification is one of the bases for diagnosis by doctors, the calcification in the image needs to be recovered after the head dose CT image is denoised. The CT image calcification point restoration after noise reduction specifically comprises the following steps: fusing the head low-dose CT image subjected to noise reduction with the original head low-dose CT image to obtain a fused CT image; data preprocessing: dividing the original head low-dose CT image, the original head high-dose CT image and the fused CT image into image blocks, and normalizing the CT values of the obtained original head low-dose CT image block and the obtained fused CT image block; constructing a second network model and initializing network parameters; training a network model II: simultaneously inputting the head low-dose CT image block and the fused CT image block into a training network in a network model II, and saving the optimal parameters of the network model II when the loss function value of the network model II reaches a set threshold value; restoring calcifications of the CT image after noise reduction: and (3) setting the parameters of the network model II as the stored optimal parameters by using the optimal parameters of the network model II, inputting the original head low-dose CT image and the fused CT image into the network model II, and outputting the network, namely the head CT image with the recovered and clear calcifications by using the network.
The invention uses the U-net network to recover calcifications for the head low-dose CT image after noise reduction, and the network structure is shown in figure 2. The U-net network of the invention comprises two inputs, namely input 1 and input 2, wherein input 1 is a head low-dose CT image, and input 2 is a fusion CT image of the head low-dose CT image and the head CT image after noise reduction. The benefit of the U-net network design for two inputs is: if only the head CT image after noise reduction is input into the U-net network (the calcifications in the CT image after noise reduction basically disappear), the U-net network cannot recover the calcifications for the CT image after noise reduction because the network cannot obtain more information of the calcifications. To solve this problem, the method is designed to add an input, input 1, to the U-net network. The input 1 is a head low-dose CT image acquired, calcification points on the image are well preserved, and the calcification point information can be acquired by the U-net network by inputting the CT image with the well preserved calcification points into the network. The input 2 of the U-net network is a fused image of a head low-dose CT image and a head CT image subjected to noise reduction, and the reason why the fused CT image instead of the head CT image subjected to noise reduction is taken as the input of the network is as follows: not only is the information of the calcifications weakened in the head CT image after noise reduction, but also some image details (except noise and calcifications) are lost on the CT image after noise reduction compared with the original head low-dose CT image. Therefore, the method of fusing the head low-dose CT image and the head CT image subjected to noise reduction is adopted, the fused CT image can restore information (such as calcifications, noise and the like) lost by the image, the restored calcifications are beneficial to extracting more calcifications by a network, and the restored noise information does not influence the noise reduction effect of the image, because the U-net network also has a certain noise reduction effect.
In the prior art, calcifications are generally identified and reserved in a clear CT image, the CT value of each pixel point on the image is firstly calculated, calcified pixel points are obtained according to a set threshold value, then the calcified pixel points are connected into blocks by using a component connection algorithm to obtain the calcifications, and finally the calcifications are marked and reserved, so that the effect that the calcifications are processed when the image with the calcifications is post-processed later is avoided. Compared with the prior art, the invention has the following beneficial effects:
(1) the invention adopts a deep learning network to reduce noise of the low-dose CT image and simultaneously reserve calcifications in the image.
(2) According to the method provided by the invention, the calcifications in the low-dose head CT image are not required to be identified firstly, but the head CT image after noise reduction is input into the trained deep learning network, and the CT image output by the network not only keeps the definition of the input image, but also recovers the calcifications in the original low-dose CT image (if the calcifications exist in the original low-dose CT image).
(3) According to the method, before the noise of the head low-dose CT image is reduced, the calcification points in the CT image are marked without manpower or a special method, so that the workload is greatly reduced.
Drawings
Fig. 1 is a structural diagram of a noise reduction network (network model one) according to the present embodiment;
FIG. 2 is a diagram of a calcification recovery network (network model II) according to the present embodiment;
FIG. 3 is a diagram illustrating a noise reduction process of a low-dose CT image of a head using a deep learning network according to the present embodiment;
FIG. 4 is a process diagram of restoring calcifications in a head CT image by using a deep learning network according to the embodiment;
FIG. 5 is a diagram illustrating the noise-reduced result of the present embodiment;
fig. 6 is a graph showing the recovery result of calcifications in this example.
Detailed Description
In order to make the technical means of the present invention and the technical effects achieved thereby clearer and more complete, an embodiment is provided, and the following detailed description is made with reference to the accompanying drawings:
a head low-dose CT image calcification spot retaining method based on deep learning comprises
S100, restoring calcifications of the CT image after noise reduction of the low-dose CT image and noise reduction of S200;
wherein S100 specifically comprises
S101, acquiring data: and obtaining an original head low-dose CT image and an original head high-dose CT image corresponding to the original head low-dose CT image.
S102, data preprocessing: and dividing the head low-dose CT image and the head high-dose CT image into image blocks, and normalizing the CT values of the head low-dose CT image block and the head high-dose CT image block obtained by division. The specific process is as follows: the size of the obtained original low-dose CT image and the high-dose CT image is N × N (the size of the original head low-dose CT image used in this embodiment is 512 × 512), the size of the image is too large, and in order to save the memory and the display memory of a computer and increase the amount of training data, the obtained original head CT image needs to be segmented, in this embodiment, the original head low-dose CT image and the high-dose CT image are segmented into image blocks of M × M size (the size of the image block can be adjusted at will, and the size of the image block used in this embodiment is 64 × 64), and then the obtained low-dose CT image block and the high-dose CT image block corresponding to the low-dose CT image block are normalized to [0, 1 ].
S103, constructing a network model I (GAN network), wherein the GAN network comprises a generating network and a judging network, and initializing network parameters of the generating network and the judging network. Specifically, a generator assisted distributed network (GAN) is used by the noise reduction network, and the generation of the GAN includes a generation network and a discrimination network, and the network structure of the GAN is shown in fig. 1. The input of the network is a head low-dose CT image, the output is a head CT image after noise reduction, namely, the network generates a high-dose CT image, the network is judged to judge whether the input image is true or false, the network generates a high-dose CT image and the high-dose CT image are input into the network, and the network generates a high-dose CT image and the probability that the high-dose CT image is true.
S104, training a first network model: inputting the head low-dose CT image block and the head high-dose CT image block obtained by segmentation into a first network model to train a network, and storing the optimal parameters of the first network model when the loss function value of the first network model reaches a set threshold value; inputting the head low-dose CT image block (represented by low, namely the actual input at this time is the head low-dose CT image block processed in the step S102) obtained by segmentation into a generation network to obtain a network generated head high-dose CT image block, inputting the head high-dose CT image block (represented by normal, namely the actual input at this time is the head high-dose CT image block processed in the step S102) obtained by segmentation and the network generated high-dose CT image block into a discrimination network, discriminating the probability that the network generated high-dose CT image block is the real image block and the probability that the high-dose CT image block is the real image block, and updating network parameters of the generation network and the discrimination network according to the obtained probability values; the steps are repeated until the network can not judge whether the network generated high-dose CT image block and the high-dose CT image block are true or false (the true probability values of the network generated high-dose CT image and the high-dose CT image are almost equal), and the parameters (namely the optimal parameters) of the network at the moment are stored.
S105, denoising the low-dose CT image: inputting an original head low-dose CT image into the network model I by using the optimal parameters of the network model I, and outputting the head low-dose CT image subjected to noise reduction; the method specifically comprises the following steps: initializing the generation network by using the generation network parameters obtained in S104, inputting the original head low-dose CT image (the image at this time does not need to be divided, and the input complete head CT image) which needs to be denoised into the generation network, and obtaining the denoised head CT image, with the result as shown in fig. 5. In the figure, low represents an original head low-dose CT image, normal represents an original head high-dose CT image corresponding to the low, and predict is a head CT image after noise reduction by using a generation network, and the low and predict can find that noise is reduced greatly in the predict, but calcifications in the predict are more fuzzy than calcifications in the low (marked by arrows).
Wherein S200 specifically comprises
S201 acquires data: fusing the head low-dose CT image subjected to noise reduction with the original head low-dose CT image to obtain a fused CT image; the method specifically comprises the following steps: using the original head low-dose CT image and the original head high-dose CT image obtained in S100, inputting the original head low-dose CT image into a training generation network, outputting to obtain a noise-reduced head CT image, and fusing the original head low-dose CT image and a noise-reduced image corresponding to the original head low-dose CT image together to obtain a fused CT image (i.e., input 2 in fig. 2, input 2= a × low + (1-a) × predict, a controls the fusion ratio of the low-dose CT image and the noise-reduced CT image, a can be obtained by actual debugging, 0< a <1, low in the formula is the original head low-dose CT image, and predict is the noise-reduced head CT image).
S202, data preprocessing: and (4) segmenting the original head low-dose CT image, the original head high-dose CT image and the fused CT image, and performing CT value normalization processing on segmented image blocks, wherein the method is the same as that in the step S102.
S203, constructing a second network model (U-Net network) and initializing network parameters.
S204, training a network model II: and simultaneously inputting the segmented head low-dose CT image and the segmented fusion CT image into a second network model for training, and storing the optimal parameters of the second network model. The method specifically comprises the following steps: inputting the head low-dose CT image block obtained by segmentation and the fused CT image block obtained by segmentation into a network, wherein the output of the network is a calcification point recovery CT image block, calculating a loss value between the calcification point recovery CT image block and the head high-dose CT image block according to a loss function of the network model II, and updating network parameters according to the loss value; and continuously calculating the loss value between the calcification point recovery CT image block and the head high-dose CT image block according to the loss function to update the network parameters until the loss value reaches a set threshold value, and storing the optimal parameters of the network.
S205, restoring calcifications of the noise-reduced CT image: and (3) setting the parameters of the second network model as the parameters stored in the s204 by using the optimal parameters of the second network model, and inputting the original head low-dose CT image and the fused CT image into the second network model to obtain the head CT image with the recovered and clear calcifications. The method specifically comprises the following steps: the method comprises the steps of firstly using a noise reduction network to reduce noise of a head low-dose CT image needing noise reduction, then fusing the head CT image subjected to noise reduction with an original head low-dose CT image to obtain a fused CT image, inputting the original head low-dose CT image and the fused CT image into a trained U-Net network simultaneously, and outputting the network to obtain a clear CT image with recovered calcifications, wherein the result is shown in figure 6. Wherein low is an original head low-dose CT image, predict is a head CT image subjected to noise reduction, output is a head CT image subjected to calcification point recovery, and normal is an original head high-dose CT image. The noise level of both the Predict image and the output image is reduced compared to low, and the calcifications in the output image are clearer than those in the Predict image (arrows).
Therefore, the calcification retaining method for head low-dose CT image reconstruction based on deep learning can retain calcification which is quite similar to noise points in an image while reducing noise of the head low-dose CT image, and plays a vital role in head low-dose CT image diagnosis.
The above description is provided for the purpose of further elaboration of the technical solutions provided in connection with the preferred embodiments of the present invention, and it should not be understood that the embodiments of the present invention are limited to the above description, and it should be understood that various simple deductions or substitutions can be made by those skilled in the art without departing from the spirit of the present invention, and all such alternatives are included in the scope of the present invention.
Claims (5)
1. A head low-dose CT image calcification spot retaining method based on deep learning comprises
S100 low-dose CT image noise reduction, specifically comprising
S101, acquiring data: obtaining an original head low-dose CT image and an original head high-dose CT image corresponding to the original head low-dose CT image;
s102, data preprocessing: dividing the head low-dose CT image and the head high-dose CT image into image blocks, and normalizing the CT values of the low-dose CT image block and the high-dose CT image block obtained by division;
s103, constructing a first network model and initializing network parameters;
s104, training a first network model: inputting the head low-dose CT image block and the head high-dose CT image block obtained by segmentation into a first network model to train a network, and storing the optimal parameters of the first network model when the loss function value of the first network model reaches a set threshold value;
s105, denoising the low-dose CT image: setting the parameters of the network model I as the parameters stored in S104 by using the optimal parameters of the network model I, inputting an original head low-dose CT image into the network model I, and outputting the original head low-dose CT image after noise reduction by using a network;
s200 restoring calcifications from the CT image after noise reduction specifically comprises
S201 acquires data: fusing the head low-dose CT image subjected to noise reduction with the original head low-dose CT image to obtain a fused CT image;
s202, data preprocessing: dividing an original head low-dose CT image, an original head high-dose CT image and a fused CT image into CT image blocks, and normalizing CT values of the divided CT image blocks;
s203, constructing a second network model and initializing network parameters;
s204, training a network model II: inputting the head low-dose CT image block obtained by segmentation and the fused CT image block obtained by segmentation into a network, wherein the output of the network is a calcification point recovery CT image block, calculating a loss value between the calcification point recovery CT image block and the head high-dose CT image block, updating network parameters according to the loss value, and storing the optimal parameters of the network until the loss value reaches a set threshold value;
s205 restoring the characteristics of the calcification points: and (3) setting the parameters of the second network model as the parameters stored in the S204 by using the optimal parameters of the second network model, inputting the original head low-dose CT image and the fused CT image into the second network model, and outputting the head CT image with the restored calcifications and low noise as a network output.
2. The deep learning-based method for preserving calcifications in low-dose head CT images as claimed in claim 1, wherein: and the first network model adopts a GAN network comprising a generating network and a judging network.
3. The deep learning-based method for preserving calcifications in low-dose head CT images as claimed in claim 2, wherein: in the step S104, firstly, inputting the head low-dose CT image block into a generating network, outputting the head low-dose CT image block to obtain a network-generated head high-dose CT image block, then simultaneously inputting the image block and the head high-dose CT image block into a discriminating network, calculating the probability that the network-generated head high-dose CT image block is a real image block and the probability that the head high-dose CT image block is a real image block according to a loss function of the discriminating network, and updating the generated network parameters and the discriminating network parameters according to the obtained probability values; and repeating the steps until the judging network cannot judge whether the head high-dose CT image block generated by the network and the head high-dose CT image block are true or false, and storing the parameters of the first model.
4. The deep learning-based method for preserving calcifications in low-dose head CT images as claimed in claim 1, wherein: and the second network model adopts a U-Net network, the U-Net network has two inputs and one output, and the two inputs are a head low-dose CT image and a fusion CT image of the head low-dose CT image and the noise-reduced head CT image respectively.
5. The deep learning-based method for preserving calcifications in low-dose head CT images as claimed in claim 4, wherein: in step S204, the determining method of the optimal parameter of the U-Net network is as follows: and calculating a loss value between the network output image block and the original head high-dose CT image block according to a loss function formula of the U-net network, and modifying network parameters according to the loss value until the loss function value reaches a set threshold value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010142001.5A CN111369463B (en) | 2020-03-04 | 2020-03-04 | Head low-dose CT image calcification point retaining method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010142001.5A CN111369463B (en) | 2020-03-04 | 2020-03-04 | Head low-dose CT image calcification point retaining method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111369463A CN111369463A (en) | 2020-07-03 |
| CN111369463B true CN111369463B (en) | 2021-02-23 |
Family
ID=71211220
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010142001.5A Active CN111369463B (en) | 2020-03-04 | 2020-03-04 | Head low-dose CT image calcification point retaining method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111369463B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392204A (en) * | 2017-07-20 | 2017-11-24 | 东北大学 | A kind of galactophore image microcalcifications automatic checkout system and method |
| CN109829896A (en) * | 2019-01-14 | 2019-05-31 | 中国科学院苏州生物医学工程技术研究所 | The micro-calcification clusters automatic testing method of digital galactophore tomography X image based on multi-domain characteristics |
| CN109949215A (en) * | 2019-03-29 | 2019-06-28 | 浙江明峰智能医疗科技有限公司 | A kind of low-dose CT image simulation method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7689017B2 (en) * | 2005-12-28 | 2010-03-30 | The General Hospital Corporation | Medical image processing |
| US11250600B2 (en) * | 2018-01-12 | 2022-02-15 | Korea Advanced Institute Of Science And Technology | Method for processing X-ray computed tomography image using neural network and apparatus therefor |
-
2020
- 2020-03-04 CN CN202010142001.5A patent/CN111369463B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392204A (en) * | 2017-07-20 | 2017-11-24 | 东北大学 | A kind of galactophore image microcalcifications automatic checkout system and method |
| CN109829896A (en) * | 2019-01-14 | 2019-05-31 | 中国科学院苏州生物医学工程技术研究所 | The micro-calcification clusters automatic testing method of digital galactophore tomography X image based on multi-domain characteristics |
| CN109949215A (en) * | 2019-03-29 | 2019-06-28 | 浙江明峰智能医疗科技有限公司 | A kind of low-dose CT image simulation method |
Non-Patent Citations (2)
| Title |
|---|
| Generative Adversarial Networks for Noise;Jelmer M. Wolterink et al;《 IEEE TRANSACTIONS ON MEDICAL IMAGING》;20171231;第36卷(第12期);第2536-2545页 * |
| 生成对抗网络在医学图像处理中的应用;陈锟 等;《生命科学仪器》;20181231;第16卷;第71-80、91页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111369463A (en) | 2020-07-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109785243B (en) | Denoising method and computer based on unregistered low-dose CT of countermeasure generation network | |
| CN112017131B (en) | CT image metal artifact removing method and device and computer readable storage medium | |
| JP2011517986A (en) | Automatic detection and accurate segmentation of abdominal aortic aneurysms | |
| CN112950737B (en) | Fundus fluorescence contrast image generation method based on deep learning | |
| CN111243087B (en) | Three-dimensional reconstruction method and device for fundus blood vessel and electronic equipment | |
| CN106530236B (en) | Medical image processing method and system | |
| CN113706409A (en) | CBCT image enhancement method and device based on artificial intelligence and storage medium | |
| US20220138936A1 (en) | Systems and methods for calcium-free computed tomography angiography | |
| CN116309647B (en) | Method for constructing craniocerebral lesion image segmentation model, image segmentation method and device | |
| CN111369463B (en) | Head low-dose CT image calcification point retaining method based on deep learning | |
| CN117409002A (en) | Visual identification detection system for wounds and detection method thereof | |
| CN117197270A (en) | Metal artifact correction method and device for nonlinear projection decomposition and imaging equipment | |
| Zhang et al. | Image restoration of medical images with streaking artifacts by Euler's elastica inpainting | |
| CN115082585A (en) | Medical image reconstruction method | |
| Noviana et al. | Axial segmentation of lungs CT scan images using canny method and morphological operation | |
| CN113570586A (en) | Method and device for creating and processing CT image of neural network system | |
| CN115775219A (en) | Medical image segmentation method, system, electronic device, and medium | |
| Zhu et al. | CT metal artifact correction assisted by the deep learning-based metal segmentation on the projection domain | |
| Budumuru et al. | Preprocessing analysis of medical image: a survey | |
| Pandey et al. | A Framework for Mathematical Methods in Medical Image Processing | |
| Li et al. | A prior-interpolation based metal artifact reduction algorithm in computed tomography | |
| CN114494498B (en) | Metal artifact removing method based on double-domain Fourier neural network | |
| EP4328859A1 (en) | Method and system for removing foreign material from images | |
| Saxena et al. | A locally adaptive edge preserving filter for denoising of low dose CT using multi-level fuzzy reasoning concept | |
| CN113628187A (en) | Automatic migratable artificial intelligence medical image recognition system and recognition method |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |