A CT image marking method, system, medium and device

Abstract

The present invention relates to the field of image processing, and in particular to a CT image marking method, system, medium and device. The method comprises: step 1, obtaining a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae; step 2, performing a de-adhesion process on the first 2D segmentation mask image set of ribs to obtain a first segmentation image set; step 3, performing a grouping process on the first segmentation image set to obtain an image data set of n groups of ribs; step 4, judging the image data set belonging to the same rib from the image data sets of n groups of ribs to obtain a judgment result; step 5, determining the 3D center line of the vertebrae according to the second 2D segmentation mask image set of the vertebrae, and completing the marking of the rib image based on the 3D center line and the judgment result. The present invention can achieve accurate marking of image data without requiring a large number of models and labor costs.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to a CT image marking method, system, medium and equipment. Background Art

[0002] Most existing rib counting methods use deep learning neural network technology to predict the number of ribs in each sequence of CT images. This solution requires a large amount of chest CT data with 24 categories of labels to support it. The cost of manual labeling is high, time-consuming, and difficult to implement. In addition, the 24 ribs are highly similar and are arranged in close proximity, which not only makes it difficult for labelers to label the categories, but also causes inaccurate neural network predictions. In addition, there are methods that perform single-category segmentation on CT ribs to obtain a binary mask, and then use image processing operations such as expansion and corrosion to solve the slight adhesion between ribs. Finally, the ribs are labeled through connected domain operations. This method is not applicable to situations where broken bones exist, and it is not applicable to situations where adjacent ribs are in close contact. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a CT image marking method, system, medium and equipment.

[0004] The technical solution of the present invention to solve the above technical problem is as follows: A CT image marking method, comprising:

[0005] Step 1, inputting the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, converting the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae;

[0006] Step 2, performing a de-adhesion process on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set;

[0007] Step 3, grouping the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set;

[0008] Step 4, judging the image data sets belonging to the same rib from the n groups of rib image data sets, and obtaining a judgment result;

[0009] Step 5: determine the 3D centerline of the vertebra according to the second 2D segmentation mask image set of the vertebra, and complete the labeling of the rib image based on the 3D centerline and the judgment result.

[0010] The beneficial effects of the present invention are as follows: conversion of 3D images can facilitate subsequent data extraction and merging processing; de-adhesion can effectively avoid the situation where two rib areas in the image are too close and thus cause deviations in the final labeling results; grouping processing of the first segmented image set can quickly synthesize two-dimensional images into complete and continuous three-dimensional images; and by judging the image data set of ribs, an image set of images from the same rib, namely, three-dimensional images, can be obtained, which is convenient for subsequent labeling of image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0011] Based on the above technical solution, the present invention can also be improved as follows.

[0012] Further, step 2 specifically includes:

[0013] Step 201, performing a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, if the first judgment result of the first rib region is an adhesion region, performing intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and updating the intersection processing result as a new first rib region, wherein i is a positive integer;

[0014] Step 202, repeating step 201 until all first rib regions on the i-th image are updated;

[0015] Step 203, repeating steps 201 and 202 until all images in the first 2D segmented image set are updated to obtain a first segmented image set;

[0016] The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

[0017] Further, step 3 specifically includes:

[0018] Step 301, performing a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determining the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region as belonging to the same rib, generating a first group of rib image data sets for the rib regions determined to be belonging to the same rib, and updating the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, generating a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not being greater than the second threshold, wherein j is a positive integer;

[0019] Step 302, repeating step 301 until the calculation of all third rib regions on the j-th image is completed;

[0020] Step 303, repeating steps 301 and 302 until the calculation of all images in the first segmented image set is completed, and n groups of rib image data sets are obtained.

[0021] Further, step 4 specifically includes:

[0022] Step 401, judging n groups of rib image data sets according to the trend characteristics of the rib shape, and obtaining the end position region of each group of rib image data sets, wherein the end position region is the rib region whose adjacent values ​​are 1 and which is the end position;

[0023] Step 402, a first tentative tentacle is extended outward from the end position area, the first tentative tentacle is a cubic block of a preset size, and it is determined whether the first tentative tentacle intersects with the first 2D segmentation mask image set of the vertebra, and a second judgment result is generated, and the image data set of the rib corresponding to the end position area where the second judgment result is yes is defined as the first main rib;

[0024] Step 403, extending a second probing tentacle outward from the end position area where the second judgment result is negative, determining whether the second probing tentacle intersects with the first main rib, and generating a third judgment result;

[0025] Step 404, defining the second probing tentacle whose third judgment result is yes and the corresponding intersecting first main rib as the second main rib;

[0026] Step 405, repeating steps 403 to 404 until the judgment of the image data sets of all ribs is completed.

[0027] Further, step 5 specifically includes:

[0028] The 3D centerline of the vertebra is determined according to the first 2D segmentation mask image set of the vertebra, the 3D mask center of each group of rib image data sets is determined according to the judgment result, and the rib image is marked based on the 3D centerline and the 3D mask center.

[0029] Another technical solution of the present invention to solve the above technical problem is as follows: A CT image marking system, comprising:

[0030] a conversion module, configured to input the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, and convert the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae;

[0031] A de-adhesion module, used for performing a de-adhesion process on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set;

[0032] A processing module, used for grouping the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set;

[0033] A judgment module, used for judging the image data sets belonging to the same rib from the n groups of rib image data sets, and obtaining a judgment result;

[0034] The marking module is used to determine the 3D center line of the vertebra according to the second 2D segmentation mask image set of the vertebra, and complete the marking of the rib image based on the 3D center line and the judgment result.

[0035] The beneficial effects of the present invention are as follows: conversion of 3D images can facilitate subsequent data extraction and merging processing; de-adhesion can effectively avoid the situation where two rib areas in the image are too close and thus cause deviations in the final labeling results; grouping processing of the first segmented image set can quickly synthesize two-dimensional images into complete and continuous three-dimensional images; and by judging the image data set of ribs, an image set of images from the same rib, namely, three-dimensional images, can be obtained, which is convenient for subsequent labeling of image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0036] Further, the de-adhesion module is specifically used for:

[0037] an intersection module, configured to perform a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, and if the first judgment result of the first rib region is an adhesion region, perform intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and update the intersection processing result as a new first rib region, wherein i is a positive integer;

[0038] A first repetition module, repeating the intersection module until the update of all first rib regions on the i-th image is completed;

[0039] A second repetition module repeats the intersection module and the first repetition module until all images in the first 2D segmented image set are updated to obtain a first segmented image set;

[0040] The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

[0041] Furthermore, the processing module is specifically used for:

[0042] an updating module, configured to perform a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determine that the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region are the same ribs, generate a first group of rib image data sets for the rib regions determined to be the same ribs, and update the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, and generate a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not greater than the second threshold, wherein j is a positive integer;

[0043] A third repetition module, used for repeating the update module until the calculation of all third rib regions on the j-th image is completed;

[0044] The fourth repetition module is used to repeat the updating module and the third repetition module until the calculation of all images in the first segmented image set is completed to obtain n groups of rib image data sets.

[0045] Further, the judgment module is specifically used for:

[0046] A position judgment module judges the n groups of rib image data sets according to the trend characteristics of the rib shape, and obtains the end position area of ​​each group of rib image data sets, wherein the end position area is the rib area with adjacent values ​​of 1 and the end position;

[0047] A setting module is provided, wherein a first tentative tentacle is extended outwardly at the end position area, wherein the first tentative tentacle is a cubic block of a preset size, and whether the first tentative tentacle intersects with a first 2D segmentation mask image set of the vertebra is determined, and a second determination result is generated, and an image data set of the rib corresponding to the end position area where the second determination result is yes is defined as a first main rib;

[0048] an expansion module, extending a second probing tentacle outward from the end position area where the second judgment result is negative, determining whether the second probing tentacle intersects with the first main rib, and generating a third judgment result;

[0049] A determination module defines the second probing tentacle with a third judgment result of yes and the corresponding intersecting first main rib as the second main rib;

[0050] The fifth repetition module repeats the expansion module to the determination module until the judgment of the image data sets of all ribs is completed.

[0051] Furthermore, the marking module is specifically used for:

[0052] The 3D centerline of the vertebra is determined according to the first 2D segmentation mask image set of the vertebra, the 3D mask center of each group of rib image data sets is determined according to the judgment result, and the rib image is marked based on the 3D centerline and the 3D mask center.

[0053] Another technical solution of the present invention to solve the above technical problem is as follows: a storage medium, in which instructions are stored. When a computer reads the instructions, the computer executes a CT image marking method as described in any one of the above.

[0054] The beneficial effects of the present invention are as follows: conversion of 3D images can facilitate subsequent data extraction and merging processing; de-adhesion can effectively avoid the situation where two rib areas in the image are too close and thus cause deviations in the final labeling results; grouping processing of the first segmented image set can quickly synthesize two-dimensional images into complete and continuous three-dimensional images; and by judging the image data set of ribs, an image set of images from the same rib, namely, three-dimensional images, can be obtained, which is convenient for subsequent labeling of image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0055] Another technical solution of the present invention to solve the above technical problem is as follows: an electronic device comprises the above storage medium and a processor for executing instructions in the above storage medium.

[0056] The beneficial effects of the present invention are as follows: conversion of 3D images can facilitate subsequent data extraction and merging processing; de-adhesion can effectively avoid the situation where two rib areas in the image are too close and thus cause deviations in the final labeling results; grouping processing of the first segmented image set can quickly synthesize two-dimensional images into complete and continuous three-dimensional images; and by judging the image data set of ribs, an image set of images from the same rib, namely, three-dimensional images, can be obtained, which is convenient for subsequent labeling of image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a flow chart of an embodiment of a CT image marking method of the present invention;

[0058] Figure 2 A structural framework diagram provided for an embodiment of a CT image marking system of the present invention. DETAILED DESCRIPTION

[0059] The principles and features of the present invention are described below. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0060] like Figure 1 As shown, a CT image marking method includes:

[0061] Step 1, inputting the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, converting the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae;

[0062] Step 2, performing a de-adhesion process on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set;

[0063] Step 3, grouping the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set;

[0064] Step 4, judging the image data sets belonging to the same rib from the n groups of rib image data sets, and obtaining a judgment result;

[0065] Step 5: determine the 3D centerline of the vertebra according to the first 2D segmentation mask image set of the vertebra, and complete the labeling of the rib image based on the 3D centerline and the judgment result.

[0066] In some possible implementations, conversion of 3D images can facilitate subsequent data extraction and merging processing, and de-adhesion can effectively avoid the situation where two rib areas in the image are too close, resulting in deviations in the final labeling results. By grouping the first segmented image set, the two-dimensional image can be quickly synthesized into a complete and continuous three-dimensional image. By judging the rib image data set, a set of image images from the same rib, namely, a three-dimensional image, can be obtained, which is convenient for subsequent labeling of the image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0067] It should be noted that the CT image set contains multiple CT images, the first 3D segmentation mask image set of the ribs contains 3D segmentation mask images of multiple ribs, the second 3D segmentation mask image set of the vertebrae contains 3D segmentation mask images of multiple vertebrae, the first 2D segmentation mask image set of the ribs contains 2D segmentation mask images of multiple ribs, the second 2D segmentation mask image set of the vertebrae contains 2D segmentation mask images of multiple vertebrae, the first segmentation image set contains 2D segmentation mask images of multiple de-adhesioned ribs, any rib mask image data set contains multiple mask images constituting the same rib, any complete bone mask image data set contains multiple mask images constituting the same complete bone, the segmentation model can be: Unet, FCN, DSN and other semantic segmentation network models, the conversion process is to convert the 3D segmentation mask image to the coronal plane perspective through a 3D matrix rotation operation.

[0068] The specific process of step 2 may be understood by referring to Example 1.

[0069] Example 1: Depending on the accuracy of the segmentation network, there are cases where the masks of two or more ribs are adhered to each other in the mask image. In addition, in reality, there may also be cases where adjacent ribs are in contact with each other, and the masks are naturally adhered. Therefore, first of all, it is necessary to remove the adhesion, that is, de-adhesion processing:

[0070] (1) Assume that the sequence number of the current 2D CT (i.e., the first 2D segmentation mask image of the rib, i.e., an image in) is i, i is initially 0, and x rib regions can be obtained on its coronal mask, and the mask of the i+1 sequence has y rib regions;

[0071] (2) Calculate whether there is a rib region a in the y regions and the Dice of multiple rib regions (b1, b2, ...) in the i sequence exceeds 0.1. If so, it means that region a is an adhesion region and proceed to step (3). Otherwise, execute i = i + 1 and proceed to step (2) again. Dice is the value obtained by comparing the segmentation mask calculated by the algorithm with the correct mask, and is a widely used indicator for evaluating segmentation accuracy.

[0072] (3) Calculate the intersection of the mask of region a and the masks of b1, b2, ... respectively as the substitute mask of region a. Then, execute i = i + 1 and repeat step (2). The process of calculating the intersection is as follows: put a and b1, b2, ... on the same plane. The area with a mask value of 1 (rib segmentation area) will have an intersecting part, that is, the intersection. The substitute mask of region a can be understood as: only retain the part of the original region a that intersects with b1, b2, ...

[0073] The specific process of step 3 may be understood by referring to Example 2.

[0074] Example 2: After removing adhesions, the rib regions are grouped (each rib is a group. It should be noted that each rib can be understood as a complete rib or a section of ribs. As long as it is a complete and unbroken section of bones, it can be a rib). The method is as follows:

[0075] (1) The traversal order is the same as a. Let j be the current sequence number, and its coronal plane has p rib regions. Traverse each region s and calculate the dice between it and each rib region t in the j-1th sequence. If dice>0.1, group s and t into one group (s and t belong to the same rib), and record the number of neighbors of s (the number of rib regions in different sequences that belong to the same group with it and whose dice is greater than 0.1), that is, the number of adjacent values. If there is no region with a dice greater than 0.1 with s, group it into a new group (a new rib appears), and N groups of data can be obtained.

[0076] The specific process of step 4 may be understood by referring to Example 3.

[0077] Example 3, the method for finding the end position of each rib (it should be noted here that each rib has two end positions) is as follows:

[0078] Each of the N groups of data represents a rib, and each group of data includes multiple 2D regions. The 2D region with neighbor of 1 represents that this rib may be the end position. According to the trend of the rib shape from the back to the chest in the coronal plane (from a certain middle area of ​​the rib to the two ends of the rib), it can be found that the neighbor of the starting middle area may also be 1, that is, by observing the CT2D sequence images from the back to the chest, it can be found that the first rib seen is a certain area between the two ends. According to the algorithm steps of this patent, it can be concluded that the neighbor of this area is also 1, and these areas can be excluded. In this way, the remaining areas with neighbor of 1 are the two end positions.

[0079] A 10 pixel * 10 pixel * 10 pixel cube is expanded outward from the end of each rib as a trial tentacle. If the tentacle intersects with the vertebrae mask, that is, the tentacle position in the image overlaps with the area with a mask value of 2 (that is, the vertebrae area), then this rib is marked as the main rib, that is, the main rib. In this way, the end of the main rib near the vertebra and the other end can also be obtained. If it does not intersect with the vertebrae mask, it is marked as a broken rib.

[0080] Extend an active tentacle of 30 pixels * 30 pixels * 30 pixels from the non-vertebral end of the main rib, and extend a passive tentacle of 30 pixels * 30 pixels * 30 pixels from the two ends of the broken rib. If there is an active tentacle intersecting with a passive tentacle, it means that the two ribs may have originally belonged to the same rib. If there are multiple passive tentacles intersecting with the active tentacles, select the pair with the largest 3dIOU of the two tentacles, and then reclassify the corresponding main rib and broken rib into the same group, so that the broken bone search is realized. IOU is a widely used calculation method, iou = intersection of two regions / union of two regions.

[0081] The specific process of step 5 may be understood by referring to Example 4.

[0082] Example 4: Calculate the center of each 2D vertebra mask area, connect these center points, and extract the 3D center line of the vertebra. The center of the 3D mask composed of each group of ribs, that is, the point with the largest distance from the edge of the area where the mask is 1 (in the rib area) is the center point of the area to distinguish the left and right ribs. The rib center is on the left side of the vertebra center line, which is the left rib. Similarly, the right rib can be obtained. Then, the upper and lower positions of each group of ribs are sorted by number, that is, the upper and lower positions are marked from the first rib to the twelfth rib. At this point, the entire rib counting and marking process is completed. The upper and lower positions of each group of ribs are based on the coordinate points in 3D space.

[0083] Preferably, in any of the above embodiments, step 2 specifically includes:

[0084] Step 201, performing a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, if the first judgment result of the first rib region is an adhesion region, performing intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and updating the intersection processing result as a new first rib region, wherein i is a positive integer;

[0085] Step 202, repeating step 201 until all first rib regions on the i-th image are updated;

[0086] Step 203, repeating steps 201 and 202 until all images in the first 2D segmented image set are updated to obtain a first segmented image set;

[0087] The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

[0088] Preferably, in any of the above embodiments, step 3 specifically includes:

[0089] Step 301, performing a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determining the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region as belonging to the same rib, generating a first group of rib image data sets for the rib regions determined to be belonging to the same rib, and updating the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, generating a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not being greater than the second threshold, wherein j is a positive integer;

[0090] Step 302, repeating step 301 until the calculation of all third rib regions on the j-th image is completed;

[0091] Step 303, repeating steps 301 and 302 until the calculation of all images in the first segmented image set is completed, and n groups of rib image data sets are obtained.

[0092] Preferably, in any of the above embodiments, step 4 specifically includes:

[0093] Step 401, judging n groups of rib image data sets according to the trend characteristics of the rib shape, and obtaining the end position region of each group of rib image data sets, wherein the end position region is the rib region whose adjacent values ​​are 1 and which is the end position;

[0094] Step 402, a first tentative tentacle is extended outward from the end position area, the first tentative tentacle is a cubic block of a preset size, and it is determined whether the first tentative tentacle intersects with the first 2D segmentation mask image set of the vertebra, and a second judgment result is generated, and the image data set of the rib corresponding to the end position area where the second judgment result is yes is defined as the first main rib;

[0095] Step 403, extending a second probing tentacle outward from the end position area where the second judgment result is negative, determining whether the second probing tentacle intersects with the first main rib, and generating a third judgment result;

[0096] Step 404, defining the second probing tentacle whose third judgment result is yes and the corresponding intersecting first main rib as the second main rib;

[0097] Step 405, repeating steps 403 to 404 until the judgment of the image data sets of all ribs is completed.

[0098] It should be noted that the same ribs are two or more bones or rib regions belonging to the same root or group of ribs, and the end positions are the two ends of a bone. Among them, this algorithm finds the end based on the lateral trend of the ribs in the coronal plane and calculates the adjacent values. It is not judged based on the size of the bone section, so it has nothing to do with the shape, and naturally will not affect the correct finding of the two ends. The adjacent values ​​are: the number of rib regions belonging to the same group in different sequences with a dice value greater than 0.1 in the area to be compared. The trend of the rib shape is the rib shape and trend obtained from the observation order from the back to the chest.

[0099] Preferably, in any of the above embodiments, step 5 specifically includes:

[0100] The 3D centerline of the vertebra is determined according to the first 2D segmentation mask image set of the vertebra, the 3D mask center of each group of rib image data sets is determined according to the judgment result, and the rib image is marked based on the 3D centerline and the 3D mask center.

[0101] It should be noted that the point in the area where the 3D mask center is 1 (in the rib area) and the point with the largest distance from the edge of the area is the center point of the area.

[0102] like Figure 2 As shown, a CT image marking system includes:

[0103] A conversion module 100 is used to input the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, and convert the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a first 2D segmentation mask image set of vertebrae;

[0104] A de-adhesion module 200 is used to perform de-adhesion processing on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set;

[0105] The processing module 300 is used to perform group processing on the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set;

[0106] A judgment module 400 is used to judge the image data sets belonging to the same rib from the n groups of rib image data sets to obtain a judgment result;

[0107] The marking module 500 is used to determine the 3D centerline of the vertebra according to the first 2D segmentation mask image set of the vertebra, and complete the marking of the rib image based on the 3D centerline and the determination result.

[0108] In some possible implementations, conversion of 3D images can facilitate subsequent data extraction and merging processing, and de-adhesion can effectively avoid the situation where two rib areas in the image are too close, resulting in deviations in the final labeling results. By grouping the first segmented image set, the two-dimensional image can be quickly synthesized into a complete and continuous three-dimensional image. By judging the rib image data set, a set of image images from the same rib, namely, a three-dimensional image, can be obtained, which is convenient for subsequent labeling of the image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0109] Preferably, in any of the above embodiments, the de-adhesion module 200 is specifically used for:

[0110] an intersection module, configured to perform a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, and if the first judgment result of the first rib region is an adhesion region, perform intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and update the intersection processing result as a new first rib region, wherein i is a positive integer;

[0111] A first repetition module, repeating the intersection module until the update of all first rib regions on the i-th image is completed;

[0112] A second repetition module repeats the intersection module and the first repetition module until all images in the first 2D segmented image set are updated to obtain a first segmented image set;

[0113] The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

[0114] Preferably, in any of the above embodiments, the processing module 300 is specifically used for:

[0115] an updating module, configured to perform a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determine that the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region are the same ribs, generate a first group of rib image data sets for the rib regions determined to be the same ribs, and update the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, and generate a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not greater than the second threshold, wherein j is a positive integer;

[0116] A third repetition module, used for repeating the update module until the calculation of all third rib regions on the j-th image is completed;

[0117] The fourth repetition module is used to repeat the updating module and the third repetition module until the calculation of all images in the first segmented image set is completed to obtain n groups of rib image data sets.

[0118] Preferably, in any of the above embodiments, the determination module 400 is specifically used for:

[0119] A position judgment module judges the n groups of rib image data sets according to the trend characteristics of the rib shape, and obtains the end position area of ​​each group of rib image data sets, wherein the end position area is the rib area with adjacent values ​​of 1 and the end position;

[0120] A setting module is provided, wherein a first tentative tentacle is extended outwardly at the end position area, wherein the first tentative tentacle is a cubic block of a preset size, and whether the first tentative tentacle intersects with a first 2D segmentation mask image set of the vertebra is determined, and a second determination result is generated, and an image data set of the rib corresponding to the end position area where the second determination result is yes is defined as a first main rib;

[0121] an expansion module, extending a second probing tentacle outward from the end position area where the second judgment result is negative, determining whether the second probing tentacle intersects with the first main rib, and generating a third judgment result;

[0122] A determination module defines the second probing tentacle with a third judgment result of yes and the corresponding intersecting first main rib as the second main rib;

[0123] The fifth repetition module repeats the expansion module to the determination module until the judgment of the image data sets of all ribs is completed.

[0124] Preferably, in any of the above embodiments, the marking module 500 is specifically used for:

[0125] The 3D centerline of the vertebra is determined according to the first 2D segmentation mask image set of the vertebra, the 3D mask center of each group of rib image data sets is determined according to the judgment result, and the rib image is marked based on the 3D centerline and the 3D mask center.

[0126] Another technical solution of the present invention to solve the above technical problem is as follows: a storage medium, in which instructions are stored. When a computer reads the instructions, the computer executes a CT image marking method as described in any one of the above.

[0127] In some possible implementations, conversion of 3D images can facilitate subsequent data extraction and merging processing, and de-adhesion can effectively avoid the situation where two rib areas in the image are too close, resulting in deviations in the final labeling results. By grouping the first segmented image set, the two-dimensional image can be quickly synthesized into a complete and continuous three-dimensional image. By judging the rib image data set, a set of image images from the same rib, namely, a three-dimensional image, can be obtained, which is convenient for subsequent labeling of the image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0128] Another technical solution of the present invention to solve the above technical problem is as follows: an electronic device comprises the above storage medium and a processor for executing instructions in the above storage medium.

[0129] In some possible implementations, conversion of 3D images can facilitate subsequent data extraction and merging processing, and de-adhesion can effectively avoid the situation where two rib areas in the image are too close, resulting in deviations in the final labeling results. By grouping the first segmented image set, the two-dimensional image can be quickly synthesized into a complete and continuous three-dimensional image. By judging the rib image data set, a set of image images from the same rib, namely, a three-dimensional image, can be obtained, which is convenient for subsequent labeling of the image images. The present invention can save a lot of training time and cost, and the image labeling finally obtained by the present invention is more accurate.

[0130] The reader should understand that in the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples without contradiction.

[0131] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the method embodiments described above are only illustrative, for example, the division of steps is only a logical function division, and there may be other division methods in actual implementation, such as multiple steps can be combined or integrated into another step, or some features can be ignored or not executed.

[0132] If the above method is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0133] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A CT image labeling method, characterized in that: include: Step 1, inputting the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, converting the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae; Step 2, performing a de-adhesion process on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set; Step 3, grouping the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set; Step 4, judging the image data sets belonging to the same rib from the n groups of rib image data sets, and obtaining a judgment result; Step 5: determine the 3D centerline of the vertebra according to the second 2D segmentation mask image set of the vertebra, and complete the labeling of the rib image based on the 3D centerline and the judgment result.

2. A CT image marking method according to claim 1, characterized in that: The step 2 specifically includes: Step 201, performing a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, if the first judgment result of the first rib region is an adhesion region, performing intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and updating the intersection processing result as a new first rib region, wherein i is a natural number; Step 202, repeating step 201 until all first rib regions on the i-th image are updated; Step 203, repeating step 201 and step 202 until all images in the first 2D segmented image set are updated to obtain a first segmented image set; The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

3. A CT image marking method according to claim 2, characterized in that: The step 3 specifically includes: Step 301, performing a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determining the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region as belonging to the same rib, generating a first group of rib image data sets for the rib regions determined to be belonging to the same rib, and updating the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, generating a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not being greater than the second threshold, wherein j is a positive integer; Step 302, repeating step 301 until the calculation of all third rib regions on the j-th image is completed; Step 303, repeating steps 301 and 302 until the calculation of all images in the first segmented image set is completed, and n groups of rib image data sets are obtained.

4. A CT image marking method according to claim 3, characterized in that: The step 4 specifically includes: Step 401, judging n groups of rib image data sets according to the trend characteristics of the rib shape, and obtaining the end position region of each group of rib image data sets, wherein the end position region is the rib region whose adjacent values ​​are 1 and which is the end position; Step 402, a first tentative tentacle is extended outward from the end position area, the first tentative tentacle is a cubic block of a preset size, and it is determined whether the first tentative tentacle intersects with the first 2D segmentation mask image set of the vertebra, and a second judgment result is generated, and the image data set of the rib corresponding to the end position area where the second judgment result is yes is defined as the first main rib; Step 403, extending a second probing tentacle outward from the end position area where the second judgment result is negative, determining whether the second probing tentacle intersects with the first main rib, and generating a third judgment result; Step 404, defining the second probing tentacle whose third judgment result is yes and the corresponding intersecting first main rib as the second main rib; Step 405, repeating steps 403 to 404 until the judgment of the image data sets of all ribs is completed.

5. A CT image marking method according to claim 4, characterized in that: The step 5 specifically includes: The 3D centerline of the vertebra is determined according to the first 2D segmentation mask image set of the vertebra, the 3D mask center of each group of rib image data sets is determined according to the judgment result, and the rib image is marked based on the 3D centerline and the 3D mask center.

6. A CT image marking system, characterized in that: include: a conversion module, configured to input the CT image set into the segmentation model to obtain a first 3D segmentation mask image set of ribs and a second 3D segmentation mask image set of vertebrae, and convert the first 3D segmentation mask image set and the second 3D segmentation mask image set to obtain a first 2D segmentation mask image set of ribs and a second 2D segmentation mask image set of vertebrae; A de-adhesion module, used for performing a de-adhesion process on the first 2D segmentation mask image set of the ribs to obtain a first segmentation image set; A processing module, used for grouping the first segmented image set to obtain n groups of rib image data sets, wherein each group of rib image data sets is represented as: any rib mask image data set or any complete bone mask image data set; A judgment module, used for judging the image data sets belonging to the same rib from the n groups of rib image data sets, and obtaining a judgment result; The marking module is used to determine the 3D center line of the vertebra according to the second 2D segmentation mask image set of the vertebra, and complete the marking of the rib image based on the 3D center line and the judgment result.

7. A CT image marking system according to claim 6, characterized in that: The de-adhesion module is specifically used for: an intersection module, configured to perform a first judgment on any first rib region on the i-th image in the first 2D segmentation mask image set of the ribs, and if the first judgment result of the first rib region is an adhesion region, perform intersection processing on the first rib region and all second rib regions on the i+1-th image respectively, and update the intersection processing result as a new first rib region, wherein i is a natural number; A first repetition module, repeating the intersection module until the update of all first rib regions on the i-th image is completed; A second repetition module repeats the intersection module and the first repetition module until all images in the first 2D segmented image set are updated to obtain a first segmented image set; The first judgment is: calculating a first Dice value of the first rib region and all second rib regions on the (i+1)th image; if the first Dice value exceeds a first threshold, the first rib region is an adhesion region.

8. A CT image marking system according to claim 7, characterized in that: The processing module is specifically used for: an updating module, configured to perform a second Dice value calculation on any third rib region on the j-th image in the first segmented image set and all fourth rib regions on the j-1-th image, determine that the fourth rib region corresponding to the second Dice value being greater than the second threshold and the third rib region are the same ribs, generate a first group of rib image data sets for the rib regions determined to be the same ribs, and update the adjacent number of values ​​of the third rib region according to the number of fourth rib regions corresponding to the second Dice value exceeding the second threshold, and generate a second group of rib image data sets for the fourth rib region corresponding to the second Dice value not greater than the second threshold, wherein j is a positive integer; A third repetition module, used for repeating the update module until the calculation of all third rib regions on the j-th image is completed; The fourth repetition module is used to repeat the updating module and the third repetition module until the calculation of all images in the first segmented image set is completed to obtain n groups of rib image data sets.

9. A computer-readable storage medium, characterized in that: The medium stores instructions, and when a computer reads the instructions, the computer executes a CT image marking method according to any one of claims 1 to 5.

10. An electronic device, characterized in that: The invention comprises the storage medium as claimed in claim 9 and a processor for executing instructions in the storage medium.

Images