Method for detecting and displaying calcified pixels in X-ray image by using artificial neural network
The generative neural network and differentiated neural network process the dispersed tissue images of breast calcification and dispersed tissues, and the processed normal tissue images are generated, which solves the misdiagnosis caused by the high absorption of breast calcification zone on X-ray images, and improves the early screening of breast cancer Check and diagnose accuracy.
Patent Information
- Application Number
- CN202380035848.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-21
- Publication Date
- 2024-12-13
AI Technical Summary
The breast calcification area is prone to missed diagnosis due to high absorption on X-ray images, which makes it difficult to screen and diagnose breast cancer in the early stage.
Generative neural networks and discriminative neural networks are used to generate processed normal tissue images through the training data set, thereby weakening the calcified pixels in the calcified dispersed tissue images and improving the detectability of the calcified region.
It effectively reduces the missed diagnosis rate of calcification areas and improves the accuracy of early screening and diagnosis of breast cancer.
Smart Images
Figure CN119137620A_ABST
Abstract
Description
Background Art
[0001] Breast cancer ranks first in the number of cancer deaths among women worldwide. In order to reduce the impact of breast cancer on women's health, early detection and effective treatment are the primary methods to improve the survival rate of patients.
[0002] Breast calcification is one of the main signs of breast cancer, and mammography is the most commonly used imaging method in the screening and diagnosis of breast cancer.
[0003] Breast calcifications are highly absorptive to X-rays and therefore often appear as bright spots on X-ray images. Small calcifications can be as small as a few pixels in diameter and have low contrast, so they can be easily missed in the high-intensity reading environment of radiologists. Summary of the invention
[0004] The present invention discloses a method, comprising: obtaining a training data set, the training data set comprising a plurality of real normal tissue images and a plurality of calcified dispersed tissue images; and configuring a computer system, the computer system using the training data set and a discriminative neural network to implement a generative neural network. The output of the generative neural network is a processed normal tissue image generated from an image input into the generative neural network. The output of the discriminative neural network is the probability that the image input into the discriminative neural network is a real normal tissue image.
[0005] In one aspect, in each of the plurality of calcification-dispersed tissue images, a region of interest including a plurality of calcification pixels and a plurality of normal tissue pixels is delineated.
[0006] In one aspect, for each of the plurality of calcification dispersed tissue images, the number of all pixels of the region of interest of the each calcification dispersed tissue image is at least 10 times the number of all calcified pixels of the region of interest of the each calcification dispersed tissue image.
[0007] In one aspect, for each of the plurality of calcification dispersed tissue images, the number of all pixels of the region of interest of the respective calcification dispersed tissue image is at least 50% of the number of all pixels of the respective calcification dispersed tissue image.
[0008] In one aspect, configuring the computer system includes: calculating a first loss function of a parameter of the computer system, the parameter representing an association between nodes of the generative neural network, the first loss function representing: (A) when the image input to the generative neural network is a true normal tissue image, the deviation between the image input to the generative neural network and a generated image output by the generative neural network, and (B) when the image input to the generative neural network is a calcified dispersed tissue image, the probability that the generated image determined by the discriminative neural network is a true normal tissue image; and when it is determined that the first termination condition is not met, adjusting the value of the parameter.
[0009] In one aspect, configuring the computer system includes: calculating a second loss function of associations between nodes of the discriminative neural network, the second loss function representing the probability that the generated image determined by the discriminative neural network is a true normal tissue image when the image input into the generative neural network is a calcification-dispersed tissue image; and adjusting the associations between nodes of the discriminative neural network when it is determined that the second termination condition is not satisfied.
[0010] In one aspect, each of the plurality of true normal tissue images and the plurality of calcification-dispersed tissue images is an X-ray image.
[0011] This article also discloses a method, including: based on the calcification dispersion tissue image, using a neural network to generate a processed normal tissue image; and based on (A) the calcification dispersion tissue image and (B) the processed normal tissue image, generating a calcification image.
[0012] In one aspect, the generating of the calcification image is based on a difference between (A) the calcification-dispersed tissue image and (B) the processed normal tissue image.
[0013] In one aspect, generating the calcification image includes subtracting the processed normal tissue image from the calcification-dispersed tissue image to obtain a residual image.
[0014] In one aspect, generating the calcification image further comprises: for each pixel of the residual image whose value is less than 0, setting the pixel value of each pixel to 0.
[0015] In one aspect, the method further includes: generating a calcification-enhanced image based on (A) the calcification-dispersed tissue image and (B) the calcification image.
[0016] In one aspect, generating the calcification-enhanced image includes: multiplying the calcification image by a coefficient to obtain a multiplied image; and adding the multiplied image to the calcification-dispersed tissue image.
[0017] In one aspect, the method further comprises filtering out single-pixel bright spots in the calcification image.
[0018] In one aspect, the method further includes generating the calcification-dispersed tissue image, which includes: creating a foreground mask for an input image; and cropping the input image based on the foreground mask to obtain the calcification-dispersed tissue image.
[0019] In one aspect, the processed normal tissue image and the calcification-dispersed tissue image have the same size.
[0020] In one aspect, for each calcified pixel of the calcified dispersed tissue image, the corresponding pixel of the processed normal tissue image is at a different brightness; and for each normal tissue pixel of the calcified dispersed tissue image, the corresponding pixel of the processed normal tissue image is at the same brightness.
[0021] In one aspect, the calcification-dispersed tissue image is an X-ray image.
[0022] Also disclosed herein is a computer program product, which includes a non-transitory computer-readable medium having instructions recorded thereon, and which implements the above method when the instructions are executed by a computer. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1A – Figure 1D An X-ray image of a human breast region is shown.
[0024] Figure 2 A flow chart outlining a method for training an artificial neural network according to an embodiment is shown.
[0025] Figure 3A – Figure 3D An X-ray image of a human breast region is shown.
[0026] Figure 4 A flow chart outlining a method for detecting and displaying calcified pixels in an X-ray image according to an embodiment is shown. DETAILED DESCRIPTION
[0027] Training of artificial neural networks for generating processed normal tissue images
[0028] definition
[0029] Normal tissue pixels in an X-ray image of a sample (e.g., a region of a human female breast) are pixels caused by X-rays interacting with normal tissue in the sample. In one embodiment, based on empirical data, a threshold may be specified, and if the brightness of a pixel is at most the specified threshold, the pixel may be considered a normal tissue pixel.
[0030] A calcified pixel in an X-ray image of a sample is a pixel caused by X-rays interacting with a calcified area in the sample. In one embodiment, a pixel may be considered a calcified pixel if the brightness of the pixel exceeds a specified threshold.
[0031] It is worth noting that calcification absorbs X-rays more than normal tissue, so it appears as a brighter point on the X-ray image. In other words, on the X-ray image, calcification pixels have higher brightness than normal tissue pixels.
[0032] The real normal tissue image is an X-ray image having normal tissue pixels but no calcification pixels. The real normal tissue image is taken by an X-ray detector.
[0033] The calcification-dispersed tissue image is an X-ray image having both normal tissue pixels and calcification pixels. The calcification-dispersed tissue image is captured by an X-ray detector.
[0034] The processed normal tissue image is an image having normal tissue pixels but no calcification pixels. Based on the calcification dispersion tissue image, the processed normal tissue image is generated by an artificial neural network.
[0035] A calcification image is an image with (A) calcified pixels and (B) remaining pixels whose brightness is well below a specified threshold and close to zero. Therefore, a calcification image shows bright calcified pixels on a dark background.
[0036] Training Dataset
[0037] In one embodiment, a training dataset may be collected from a hospital; the training dataset may include: (A) multiple real normal tissue images of a human breast region and (B) multiple calcified scattered tissue images of a human breast region.
[0038] To illustrate, Figure 1A The actual normal tissue image 110 of the training data set is shown. It is noteworthy that the actual normal tissue image 110 only has normal tissue pixels and no calcification pixels.
[0039] To illustrate, Figure 1B The calcification-dispersed tissue image 120 of the training data set is shown. It is noteworthy that the calcification-dispersed tissue image 120 contains both normal tissue pixels and calcification pixels (eg, calcification pixels 123).
[0040] In one embodiment, the training data set may be obtained by (ie, sent to or loaded into) a computer system (not shown).
[0041] Artificial Neural Networks
[0042] In one embodiment, the generative neural network and the discriminative neural network may be constructed and trained by a computer system using a training data set.
[0043] In other words, the computer system is configured to implement a generative neural network using the training data set and the discriminative neural network.
[0044] Generative Neural Networks
[0045] In one embodiment, after being trained by a computer system (i.e., once implemented by a computer system), the generative neural network can be used to (A) receive a calcified dispersed tissue image of a human breast region as input, and (B) generate a processed normal tissue image based on the received calcified dispersed tissue image as output.
[0046] For illustration, refer to Figure 1C and Figure 1D , after being trained, the generative neural network can receive the calcification dispersion tissue image 130 ( Figure 1C ) as input, and generates a processed normal tissue image 140 based on the calcification dispersion tissue image 130 ( Figure 1D ) as output. From the appearance, all calcified pixels (e.g., calcified pixels 133) of the calcified dispersed tissue image 130 are almost replaced by pixels similar to adjacent normal tissue pixels, thereby generating a processed normal tissue image 140 ( Figure 1D ).
[0047] Discriminative Neural Networks
[0048] In one embodiment, after being trained by a computer system, a discriminative neural network may (A) receive an input image as input, and (B) generate as output a probability that the input image is a true normal tissue image.
[0049] Flowchart of neural network training
[0050] Figure 2 A flowchart 200 is shown outlining a method for training a generative neural network and a discriminative neural network using a training dataset, according to one embodiment.
[0051] In step 210, the method may include obtaining a training data set, the training data set including a plurality of real normal tissue images and a plurality of calcified dispersed tissue images. For example, in the above embodiment, the computer system obtains a training data set, the training data set including: (A) a plurality of real normal tissue images and (B) a plurality of calcified dispersed tissue images.
[0052] In step 220, the method may include configuring a computer system that implements a generative neural network using a training data set and a discriminative neural network. For example, in the above embodiment, the computer system uses the training data set to construct and train the generative neural network and the discriminative neural network.
[0053] In addition, in step 220, the output of the generative neural network is a processed normal tissue image generated from the image input into the generative neural network. For example, in the above embodiment, the generative neural network (A) receives a calcification-dispersed tissue image of a human breast region as input, and (B) generates a processed normal tissue image as output based on the received calcification-dispersed tissue image.
[0054] Additionally, in step 220, the output of the discriminative neural network is the probability that the image input to the discriminative neural network is a true normal tissue image. For example, in the above-described embodiment, the discriminative neural network (A) receives an input image as input, and (B) generates as output the probability that the input image is a true normal tissue image.
[0055] Other embodiments of network training
[0056] Annotation by region outlining
[0057] In one embodiment, in each of the multiple calcified dispersed tissue images in the training data set, a region of interest may be outlined, the region of interest including multiple calcified pixels and multiple normal tissue pixels. Figure 1B In the calcification-dispersed tissue image 120, a polygon 125 may be used to outline a region of interest 127. The region of interest 127 includes a plurality of calcification pixels (eg, calcification pixels 123) and a plurality of normal tissue pixels.
[0058] In one embodiment, for each calcified dispersed tissue image in the plurality of calcified dispersed tissue images in the training data set, the region of interest of each calcified dispersed tissue image does not necessarily include all calcified pixels of each calcified dispersed tissue image. Figure 1B , the region of interest 127 of the calcification-dispersed tissue image 120 does not include the calcified pixels 129 .
[0059] In one embodiment, before sending the training data set to a computer system for training the generative neural network and the discriminative neural network, a radiologist or other professional may manually perform region delineation (e.g., drawing a region delineation diagram) for each of the plurality of calcified scattered tissue images of the training data set. Figure 1B Polygon 125).
[0060] There are more normal tissue pixels than calcification pixels in the region of interest
[0061] In one embodiment, for each calcified dispersed tissue image in the multiple calcified dispersed tissue images of the training data set, the number of all pixels in the region of interest of each calcified dispersed tissue image may be at least 10 times the number of all calcified pixels in the region of interest of each calcified dispersed tissue image.
[0062] For example, refer to Figure 1B , for the calcified dispersed tissue image 120 of the training data set, the number of all pixels of the region of interest 127 may be at least 10 times the number of all calcified pixels (including the calcified pixels 123) of the region of interest 127. In other words, the number of all normal tissue pixels of the region of interest 127 is at least 9 times the number of all calcified pixels of the region of interest 127.
[0063] The region of interest is at least 50% of the image
[0064] In one embodiment, for each calcified dispersed tissue image in the plurality of calcified dispersed tissue images of the training data set, the number of all pixels of the region of interest of each calcified dispersed tissue image may be at least 50% of the number of all pixels of each calcified dispersed tissue image.
[0065] For example, refer to Figure 1B For the calcification-dispersed tissue image 120 of the training data set, the number of all pixels of the region of interest 127 may be at least 50% of the number of all pixels of the calcification-dispersed tissue image 120 .
[0066] The first loss function for training generative neural networks
[0067] In one embodiment, a first loss function for training a generative neural network may be calculated based on two components: (A) when the image input to the generative neural network is a true normal tissue image, the deviation between the image input to the generative neural network and a generated image output by the generative neural network based on the input image, and (B) when the image input to the generative neural network is a calcified dispersed tissue image, the probability that the generated image determined by the discriminative neural network is a true normal tissue image.
[0068] It is worth noting that during the training of the generative neural network, the image output by the generative neural network may include calcified pixels and is therefore referred to as a "generated image" rather than a "processed normal tissue image". After training, the output of the generative neural network no longer includes calcified pixels and can therefore be referred to as a "processed normal tissue image". It is worth noting that, by definition (as described above), a processed normal tissue image does not include calcified pixels.
[0069] In one embodiment, the above component (A) can be calculated using the following formula:
[0070]
[0071] In one embodiment, the above component (B) can be calculated using the following formula:
[0072]
[0073] Therefore, in one embodiment, the following formula may be used to calculate the first loss function F1:
[0074] F1=λ*L1(G)+L cGAN (G,D)
[0075] Wherein, in the above formula, λ is a model hyperparameter defined empirically, and δ is a judgment function.
[0076] In one embodiment, the training of the generative neural network may include: when it is determined that the first termination condition is not satisfied, adjusting parameter values representing associations between nodes of the generative neural network.
[0077] In one embodiment, the first termination condition may include one or more scenarios selected from the following scenarios: minimization of the first loss function; maximization of the first loss function; reaching a certain number of iterations; reaching a value of the first loss function equal to or exceeding a certain threshold; reaching a certain computing time; and / or reaching a value of the first loss function within an acceptable error limit.
[0078] Second loss function for training discriminative neural networks
[0079] In one embodiment, when the image input to the generative neural network is a calcification-dispersed tissue image, a second loss function for training the discriminative neural network may be calculated based on the probability that the generated image determined by the discriminative neural network is a true normal tissue image.
[0080] In one embodiment, the probability may be calculated using the following formula:
[0081]
[0082] Therefore, in one embodiment, the second loss function F2 may be calculated using the following formula:
[0083] F2=L cGAN (G,D)
[0084] In one embodiment, the training of the discriminative neural network may include: when it is determined that the second termination condition is not satisfied, adjusting associations between nodes of the discriminative neural network.
[0085] In one embodiment, the second termination condition may include one or more scenarios selected from the following scenarios: minimization of the second loss function; maximization of the second loss function; reaching a certain number of iterations; reaching a value of the second loss function equal to or exceeding a certain threshold; reaching a certain computing time; and / or reaching a value of the second loss function within an acceptable error limit.
[0086] Alternative embodiments of neural network training
[0087] In the above embodiment, the computer system (A) obtains a training data set, and (B) trains a generative neural network and a discriminative neural network. Alternatively, a first computer system may obtain a training data set; and a second computer system may use the training data set to train a generative neural network and a discriminative neural network, and the second computer system may receive the training data set from the first computer system.
[0088] Detection and display of calcified pixels in images of calcified scattered tissue
[0089] In one embodiment, referring to Figure 3A and Figure 3B , after training as described above, the generative neural network can be used to generate a calcified scattered tissue image 310 ( Figure 3A ) generates a processed normal tissue image 320 ( Figure 3B ).
[0090] It is worth noting that the processed normal tissue image 320 ( Figure 3B ) and the calcification dispersion tissue image 310 ( Figure 3A ) of the calcified pixels (bright spots) are normal tissue pixels. From the appearance, the calcified scattered tissue image 310 ( Figure 3A ) appear almost as if the calcified pixels are replaced by pixels similar to the adjacent normal tissue pixels, thereby producing a processed normal tissue image 320 ( Figure 3B ).
[0091] Next, in one embodiment, referring to Figure 3C , can be based on (A) calcification dispersion tissue image 310 ( Figure 3A) and (B) processed normal tissue image 320 ( Figure 3B ) generates a calcification image 330. From the appearance, the calcification dispersion tissue image 310 ( Figure 3A ) almost seem to remain unchanged, while the remaining pixels become black, thereby forming a calcification image 330 ( Figure 3C ).
[0092] Flowchart for detecting and displaying calcified pixels
[0093] Figure 4 A flow chart 400 outlining a method for detecting and displaying calcified pixels in a calcified dispersed tissue image is shown.
[0094] In step 410, the method may include: generating a processed normal tissue image using a neural network based on the calcified scattered tissue image. Figure 3A and Figure 3B , the generative neural network is based on the calcification dispersion tissue image 310 ( Figure 3A ) generates a processed normal tissue image 320 ( Figure 3B ).
[0095] In step 420, the method may include: generating a calcification image based on (A) the calcification dispersion tissue image and (B) the processed normal tissue image. Figure 3A-3C , based on (A) calcification dispersion tissue image 310 ( Figure 3A ) and (B) processed normal tissue image 320 ( Figure 3B ) generates a calcification image 330 ( Figure 3C ).
[0096] Other embodiments of calcification display
[0097] Details on how the calcification images were generated
[0098] In one embodiment, referring to Figure 3A-3C , can be based on (A) calcification dispersion tissue image 310 ( Figure 3A ) and (B) processed normal tissue image 320 ( Figure 3B ) generates a calcification image 330 ( Figure 3C ).
[0099] In one embodiment, referring to Figure 3A-3C , calcification image 330( Figure 3C ) may include generating a calcification scattered tissue image 310 ( Figure 3A ) minus the processed normal tissue image 320 ( Figure 3B ), thereby obtaining a residual image (not shown).
[0100] Then, in one embodiment, for each pixel of the residual image whose value is less than 0, the pixel value of the pixel may be set to 0.
[0101] In one embodiment, referring to Figure 3C , single-pixel bright spots in the calcification image 330 can be filtered out.
[0102] Calcification enhanced images
[0103] In one embodiment, referring to Figure 3A , Figure 3C and Figure 3D , can be based on (A) calcification dispersion tissue image 310 ( Figure 3A ) and (B) calcification image 330 ( Figure 3C ), generating a calcification enhanced image 340 ( Figure 3D ).
[0104] In one embodiment, the calcification enhanced image 340 ( Figure 3D ) may include generating a calcification image 330 ( Figure 3C ) is multiplied by a coefficient to obtain a multiplied image (not shown); the multiplied image is then added (pixel by pixel) to the calcified dispersed tissue image 310 ( Figure 3A ).
[0105] Generation of images of tissue with scattered calcifications
[0106] In one embodiment, referring to Figure 3A The calcification-dispersed tissue image 310 may be generated by: (A) creating a foreground mask for an input image; and (B) cropping the input image based on the foreground mask to obtain the calcification-dispersed tissue image 310 .
[0107] Calcification scattered tissue image compared with normal tissue image after processing
[0108] In one embodiment, the calcification-dispersed tissue image 310 ( Figure 3A ) and the processed normal tissue image 320 ( Figure 3B ) can have the same size.
[0109] In one embodiment, for the calcification-dispersed tissue image 310 ( Figure 3A ) of each calcified pixel, the processed normal tissue image 320 ( Figure 3B ) may be at a different (eg, lower) brightness. Alternatively, the processed normal tissue image 320 ( Figure 3B ) and the calcification dispersion tissue image 310 ( Figure 3A) can be obtained from the processed normal tissue image 320 ( Figure 3B ) are inferred from the normal tissue pixels near each pixel.
[0110] In one embodiment, for the calcification-dispersed tissue image 310 ( Figure 3A ) of each normal tissue pixel, the processed normal tissue image 320 ( Figure 3B ) can be at the same brightness. Alternatively, the processed normal tissue image 320 ( Figure 3B ) and the calcification dispersion tissue image 310 ( Figure 3A ) can be very similar (e.g., within 5% of the full dynamic range) to all pixels corresponding to normal tissue pixels in the image.
[0111] In one embodiment, the calcification-dispersed tissue image 310 ( Figure 3A ) can be an X-ray image.
[0112] Although various aspects and embodiments are disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for illustrative purposes only and are not intended to be limiting, with the true scope and spirit being indicated by the appended claims.
Claims
1. A method comprising: Obtaining a training data set, wherein the training data set includes a plurality of real normal tissue images and a plurality of calcified dispersed tissue images; as well as configuring a computer system that implements a generative neural network using the training data set and the discriminative neural network, wherein the output of the generative neural network is a processed normal tissue image generated from the image input into the generative neural network, and The output of the discriminative neural network is the probability that the image input into the discriminative neural network is a true normal tissue image.
2. The method according to claim 1, wherein: In each of the plurality of calcification-dispersed tissue images, a region of interest including a plurality of calcification pixels and a plurality of normal tissue pixels is outlined.
3. The method according to claim 2, wherein: For each of the plurality of calcification-distributed tissue images, the number of all pixels of the region of interest of the respective calcification-distributed tissue image is at least 10 times the number of all calcified pixels of the region of interest of the respective calcification-distributed tissue image.
4. The method according to claim 2, wherein: For each calcification-dispersed tissue image of the plurality of calcification-dispersed tissue images, the number of all pixels of the region of interest of the respective calcification-dispersed tissue image is at least 50% of the number of all pixels of the respective calcification-dispersed tissue image.
5. The method according to claim 1, wherein: The configuring the computer system comprises: Calculating a first loss function of parameters of the computer system, the parameters representing associations between nodes of the generative neural network, the first loss function representing: (A) when the image input to the generative neural network is a true normal tissue image, the deviation between the image input to the generative neural network and a generated image output by the generative neural network, and (B) when the image input to the generative neural network is a calcified dispersed tissue image, the probability that the generated image determined by the discriminative neural network is a true normal tissue image; and When it is determined that the first termination condition is not satisfied, the value of the parameter is adjusted.
6. The method according to claim 5, wherein: The configuring the computer system comprises: Calculating a second loss function of associations between nodes of the discriminative neural network, the second loss function representing the probability that the generated image determined by the discriminative neural network is a true normal tissue image when the image input into the generative neural network is a calcification-dispersed tissue image; and When it is determined that the second termination condition is not satisfied, the associations between the nodes of the discriminative neural network are adjusted.
7. The method according to claim 1, wherein: Each of the plurality of true normal tissue images and the plurality of calcification-dispersed tissue images is an X-ray image.
8. A computer program product, comprising a non-transitory computer-readable medium having instructions recorded thereon, which implement the method of any one of claims 1 to 7 when a computer executes the instructions.
9. A method comprising: Based on the calcification dispersion tissue image, a neural network is used to generate a processed normal tissue image; as well as A calcification image is generated based on (A) the calcification-dispersed tissue image and (B) the processed normal tissue image.
10. The method according to claim 9, wherein: The generating of the calcification image is based on the difference between (A) the calcification-dispersed tissue image and (B) the processed normal tissue image.
11. The method according to claim 10, wherein: The generating of the calcification image includes: subtracting the processed normal tissue image from the calcification-dispersed tissue image, thereby obtaining a residual image.
12. The method according to claim 11, wherein: The generating of the calcification image further includes: for each pixel of the residual image whose value is less than 0, setting the pixel value of each pixel to 0.
13. The method according to claim 9, further comprising: A calcification-enhanced image is generated based on (A) the calcification-dispersed tissue image and (B) the calcification image.
14. The method according to claim 13, wherein: The generating of the calcification enhanced image comprises: multiplying the calcification image by a coefficient to obtain a multiplied image; and The multiplied image is added to the calcification-dispersed tissue image.
15. The method of claim 9, further comprising filtering out single-pixel bright spots in the calcification image.
16. The method according to claim 9, further comprising generating the calcification-dispersed tissue image, wherein: The generating the calcification dispersion tissue image comprises: creating a foreground mask for the input image; and The input image is cropped based on the foreground mask to obtain the calcified dispersed tissue image.
17. The method according to claim 9, wherein: The processed normal tissue image and the calcification-dispersed tissue image have the same size.
18. The method according to claim 9, in, For each calcified pixel of the calcified dispersed tissue image, the corresponding pixel of the processed normal tissue image is at a different brightness, and Wherein, for each normal tissue pixel of the calcified dispersed tissue image, the corresponding pixel of the processed normal tissue image is at the same brightness.
19. The method according to claim 9, wherein: The calcification-dispersed tissue image is an X-ray image.
20. A computer program product, comprising a non-transitory computer-readable medium having instructions recorded thereon, which when executed by a computer implement the method of any one of claims 9 to 19.