Medical image data rapid marking method based on superpixels

Abstract

The invention discloses a medical image data rapid marking method based on superpixels, which realizes a semi-automatic universal marking scheme and solves the problems of quality consistency, target universality, manual workload, program calculation efficiency, calculation amount and the like in medical image visceral organ tissue marking.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and in particular relates to a medical image data rapid labeling method based on superpixels. Background Art

[0002] Medical imaging is an important component of medical big data, accounting for 90% of hospital data. With the rise and development of AI technology, artificial intelligence research in medical imaging has become a very popular research field. The labeling of organs and diseases is the basis of medical imaging artificial intelligence. Both machine learning and deep learning require large sample data and a large number of labels, while traditional labeling methods are time-consuming and labor-intensive, and it is difficult to ensure quality control requirements. In addition, medical imaging is affected by differences in equipment and scanning parameters, resulting in large differences in image grayscale, which also causes universal problems for automated labeling algorithms. Therefore, a more effective and universal method is needed to solve this problem, improving labeling efficiency while also improving labeling quality.

[0003] The current mainstream target contour marking methods mainly include the following methods: manual stroke-based method, region growing-based method, and artificial intelligence-based method.

[0004] The manual stroke-based method mainly uses a brush to draw a closed curve along the target outline, and the image area surrounded by the curve is the target area.

[0005] The region growing method first manually selects the seed point of the target area, finds the contour boundary based on the grayscale and gradient differences between the seed point and its surrounding pixels, obtains the local area, and repeats this operation until the overall target is grown.

[0006] The automatic segmentation method based on artificial intelligence can automatically complete the segmentation of the specified target contour mark through the model through preliminary training of the segmentation model in the actual use stage.

[0007] The disadvantages of the current methods are as follows:

[0008] The manual tracing method requires a lot of effort to complete the marking task. The operation is complicated, the workload is large, and it is time-consuming. The marking contour varies from person to person and is also affected by factors such as the display and the window width and window position display. Therefore, this method has the lowest technical content, but most people are reluctant to use it.

[0009] The method based on region growing is more convenient than the manual outlining method. It can complete the marking by adding some seed points one by one in the target area. However, it is difficult to complete the marking of the complete outline of the target area under uniform growth conditions due to the influence of target contrast, color uniformity, contour clarity, and foreground and background differences. It is necessary to constantly adjust the growth parameters, so it requires certain graphics and image experience. At the same time, there are also cases where some fuzzy boundaries cannot be grown at all, and they need to be repaired manually.

[0010] The artificial intelligence method is fully automatic segmentation, which is a relatively new labeling method that can quickly complete the segmentation of the entire target. However, the AI ​​model is not universal and can only achieve automatic segmentation for trained targets. Otherwise, it is not supported. In order to train the segmentation model, a large number of samples need to be labeled through non-artificial intelligence methods in the early stage. The segmentation accuracy is greatly affected by the model algorithm and generally requires manual refinement. Therefore, there are still few effective models at present.

[0011] Therefore, it is necessary to develop a new type of fast annotation method. Summary of the invention

[0012] The purpose of the present invention is to overcome the deficiencies of the above-mentioned prior art and provide a method for rapid labeling of medical image data based on superpixels.

[0013] In order to solve the technical problem, the technical method of the present invention is: a medical image data fast labeling method based on superpixels, comprising the following steps:

[0014] Step 1: Read in Dicom medical images;

[0015] Step 2: Grayscale mapping is performed on the Dicom medical image. There are three cases after grayscale mapping. If superpixel calculation is performed directly, jump to step 6. If the rectangular frame is manually marked, jump to step 5. If the rough inner contour of the target is manually marked, jump to step 3.

[0016] Step 3: Manually mark the approximate inner contour of the target, offsetting the true contour by the average radius of one superpixel block, but it must be ensured to be within the target area;

[0017] The offset of the true contour is the average radius of a superpixel block pixel (the average radius of a single superpixel block area is offset from the true contour). For example, each superpixel block is an irregular shape and may cover a radius within a certain range of values. Therefore, the average value is taken to ensure that the marked area does not shift too much, resulting in the tracking area being too large or too small.

[0018] Step 4: According to the inner contour manually marked in step 3, obtain the maximum and minimum values ​​of the horizontal and vertical coordinates of the inner contour, determine the minimum circumscribed rectangle based on this value, and on this basis, expand the size of 3 to 5 super pixel blocks to generate an outer frame, and jump to step 7;

[0019] Step 5: Manually mark the outer rectangular box containing the target. Mark the rectangular box according to the target size and jump to step 7.

[0020] Step 6: No marking, directly use the entire Dicom medical image as the outer frame;

[0021] Step 7: Extract the image sub-image corresponding to the outer frame obtained in steps 4, 5, and 6, calculate the super-pixel, and obtain multiple super-pixel blocks;

[0022] Step 8: Superpixel fine-tuning: This fine-tuning optimizes the matching degree of the segmentation contour by selecting one of two methods: one is adjusting the window width and window position, and the other is adjusting the number of superpixels;

[0023] Step 9: Based on the superpixel calculation results, if the inner contour is manually marked in the previous step, go to step 10, otherwise go to step 11;

[0024] Step 10: Perform an intersection operation on the manually marked inner contour and the superpixel block, calibrate the superpixel blocks that the contour passes through, and the annular outer boundary surrounded by these calibrated superpixel blocks is the target contour. All superpixel blocks in the contour are set as the target area through the hole filling algorithm to obtain the marking result, and then go to step 11;

[0025] Step 11: Manually select superpixel blocks or delineate superpixel blocks in the contour area and automatically fill them, or select and correct the deviated superpixel blocks in step 10, and finally obtain the effective target area and convert it into a binary image. The inner and outer contours of this binary image are mapped to the original Dicom medical image as the labeling result.

[0026] Preferably, step 2 performs grayscale mapping on the original Dicom medical image according to the default window width and window position set for the examination item, so that the grayscale values ​​of the Dicom images obtained by different devices and different scanning parameters are controlled within a unified standard range.

[0027] Preferably, the grayscale mapping method is as follows: first, a standard image is determined from the sample set, and the window width and window position of the Dicom image are obtained according to the Dicom fields Window Center and Window Width. ww and g wc , based on the window width and window position, the effective visible grayscale range is:

[0028]

[0029] Then, for any Dicom image, the corresponding effective visible grayscale range can be obtained as g' min and g' max ; Let a gray value of the Dicom image be G' org , then the grayscale mapping calculation formula is as follows:

[0030]

[0031] Where:

[0032] G' org is the CT value of the original Dicom image;

[0033] G' new It is the grayscale value after the window width and window position are adjusted.

[0034] Preferably, the superpixel block in step 3 refers to the average size of all superpixel blocks, and the evaluation method is as follows: assuming that the number of superpixels is n, the size of the original Dicom image is width w, height h, then the average number of superpixel block pixels P is:

[0035]

[0036] Each superpixel is evaluated as a circle-like object, and the average radius of the superpixel is:

[0037]

[0038] As the number of superpixels increases, the average superpixel radius becomes smaller and the degree of refinement becomes higher.

[0039] Preferably, the superpixel calculation in step 7 adopts SEEDS or LSC superpixel algorithm.

[0040] Preferably, the superpixel fine-tuning in step 8 is achieved by adjusting the number of superpixels. By controlling the parameters of the superpixel block size, different numbers of superpixels are selected for target tissues of different scales or different texture complexities to control the segmentation fineness and edge matching, and the CT image is controlled within the range of [500,6000].

[0041] Preferably, the step 10 of performing the intersection operation between the manually marked inner contour and the super pixel block is as follows: recording each pixel point of the manually marked inner contour line and setting it as a set U label , generate the contour binary image F label , where the contour value is 1 and the background is 0; similarly, each superpixel block generates a binary image The method to determine whether the superpixel and the contour intersect is as follows: By traversing all superpixel blocks, all intersection relationships with the contour line are obtained, and the intersecting superpixel blocks are merged to obtain a superpixel block ring, whose outer boundary is the target contour.

[0042] Compared with the prior art, the advantages of the present invention are:

[0043] (1) The present invention discloses a method for rapid labeling of medical image data based on superpixels, which realizes a semi-automatic universal labeling scheme and solves many problems in the labeling of medical image organs and tissues, such as quality consistency, target universality, manual workload, program calculation efficiency and calculation amount;

[0044] (2) The present invention cuts the image into blocks through dynamic superpixel calculation, obtains the effective target area by matching the manually marked inner contour with the calculated superpixel block, and increases or decreases the superpixel block by manually clicking or sliding the mouse to match the target contour, thereby reducing the amount of calculation and improving the marking speed;

[0045] (3) The present invention manually marks the outer rectangular frame containing the target, which can be large or small, which can reduce the amount of superpixel calculation and the influence of the surrounding complex structure and grayscale on the superpixel calculation of the target area, thereby improving the marking speed;

[0046] (4) The manual operation part in the present invention only requires manual rough marking of the inner contour of the target, and there is no requirement for contour accuracy. The marking speed is fast and the randomness is strong, which can ensure the consistency of marking results of different users and improve marking accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 , a flowchart of a method for rapid labeling of medical image data based on superpixels of the present invention;

[0048] Figure 2 , the original CT axial slice image and its window width and window position adjustment illustration of Example 6 of the present invention;

[0049] Figure 3 , the manually drawn target outer frame diagram of embodiment 6 of the present invention;

[0050] Figure 4 , manually drawing the inner contour of the target and automatically generating the outer frame diagram according to embodiment 6 of the present invention;

[0051] Figure 5 , a target ROI image based on manual contour interception according to Embodiment 6 of the present invention;

[0052] Figure 6 , a graph showing the result of superpixel calculation of the entire image according to Embodiment 6 of the present invention;

[0053] Figure 7 , a diagram showing the ROI superpixel calculation result of Example 6 of the present invention;

[0054] Figure 8 , the ROI edge superpixel block diagram of embodiment 6 of the present invention;

[0055] Fig. 9 , the ROI target contour map of Example 6 of the present invention;

[0056] Fig.10 , target contour original image mapping diagram of Example 6 of the present invention. DETAILED DESCRIPTION

[0057] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0058] It should be noted that the contents shown in the drawings of this specification are only used to cooperate with the principles, features and content descriptions disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions under which the present invention can be implemented. Without affecting the effects and purposes that can be achieved by the present invention, they should still fall within the scope of the technical contents disclosed in the present invention.

[0059] A superpixel is a small area composed of a series of adjacent pixels with similar characteristics such as color, brightness, and texture. Most of these small areas retain effective information for further image segmentation and generally do not destroy the boundary information of objects in the image. Superpixels divide a pixel-level image into district-level images, which is an abstraction of basic information elements.

[0060] Example 1

[0061] The present invention discloses a method for rapid labeling of medical image data based on superpixels, comprising the following steps:

[0062] Step 1: Read in Dicom medical images;

[0063] Step 2: Grayscale mapping is performed on the Dicom medical image. There are three cases after grayscale mapping. If superpixel calculation is performed directly, jump to step 6. If the rectangular frame is manually marked, jump to step 5. If the rough inner contour of the target is manually marked, jump to step 3.

[0064] Step 3: Manually mark the approximate inner contour of the target, offsetting the true contour by the average radius of one superpixel block, but it must be ensured to be within the target area;

[0065] Manually mark the rough inner contour of the target. This mark does not need to be accurate. The offset size only needs to be smaller than a single superpixel block, ranging from a few pixels to more than twenty pixels. Therefore, the mark can be more casual and fast.

[0066] Step 4: According to the inner contour manually marked in step 3, obtain the maximum and minimum values ​​of the horizontal and vertical coordinates of the inner contour, determine the minimum circumscribed rectangle based on this value, and on this basis, expand the size of 3 to 5 super pixel blocks to generate an outer frame, and jump to step 7;

[0067] Step 5: Manually mark the outer rectangular box containing the target, mark the rectangular outer box according to the target size, and jump to step 7; the main purpose is to reduce the amount of superpixel calculation and the influence of the surrounding complex structure and grayscale on the superpixel calculation of the target area;

[0068] Step 6: No marking, directly use the entire Dicom medical image as the outer frame;

[0069] Step 7: Extract the image sub-image corresponding to the outer frame obtained in steps 4, 5, and 6, calculate superpixels, and obtain multiple superpixel blocks. The superpixel block is a result block calculated by the superpixel algorithm;

[0070] The meaning of image sub-image is to intercept the corresponding ROI area based on the marked outer frame as a small image containing only the target area, in order to reduce the amount of calculation and improve the marking speed.

[0071] Step 8: Superpixel fine-tuning: This fine-tuning optimizes the matching degree of the segmentation contour by selecting one of two methods: one is adjusting the window width and window position, and the other is adjusting the number of superpixels;

[0072] Step 9: Based on the superpixel calculation results, if the inner contour is manually marked in the previous step, go to step 10, otherwise go to step 11;

[0073] Step 10: Perform an intersection operation on the manually marked inner contour and the superpixel block, calibrate the superpixel blocks that the contour passes through, and the annular outer boundary surrounded by these calibrated superpixel blocks is the target contour. All superpixel blocks in the contour are set as the target area through the hole filling algorithm to obtain the marking result, and then go to step 11;

[0074] Step 11: Manually select superpixel blocks or delineate superpixel blocks in the contour area and automatically fill them, or select and correct the deviated superpixel blocks in step 10, and finally obtain the effective target area and convert it into a binary image. The inner and outer contours of this binary image are mapped to the original Dicom medical image as the labeling result.

[0075] Example 2

[0076] like Figure 1 As shown, the present invention discloses a method for rapid labeling of medical image data based on superpixels, comprising the following steps:

[0077] S1: read Dicom medical images;

[0078] S2: grayscale mapping;

[0079] S3: mark the general outline;

[0080] S4: Generate outer frame;

[0081] S5: mark the outer frame;

[0082] S6: no marker;

[0083] S7: superpixel calculation;

[0084] S8: super pixel fine-tuning;

[0085] S9: whether there is a contour mark;

[0086] S10: Calculate and match super pixel blocks according to the contour;

[0087] S11: Manual target block calibration / correction to generate a binary image.

[0088] Example 3

[0089] Preferably, step 2 performs grayscale mapping on the original Dicom medical image according to the default window width and window position set for the examination item, so that the grayscale values ​​of the Dicom images obtained by different devices and different scanning parameters are controlled within a unified standard range.

[0090] The main purpose of this step is to correct the grayscale values ​​of different images to a uniform range through grayscale mapping, as the grayscale ranges of images scanned from different sources or different devices may vary greatly.

[0091] Preferably, the grayscale mapping method is as follows: first, a standard image is determined from the sample set, and the window width and window position of the Dicom image are obtained according to the Dicom fields Window Center and Window Width. ww and g wc , based on the window width and window position, the effective visible grayscale range is:

[0092]

[0093] Then, for any Dicom image, the corresponding effective visible grayscale range can be obtained as g' min and g' max ; Let a gray value of the Dicom image be G' org , then the grayscale mapping calculation formula is as follows:

[0094]

[0095] Where:

[0096] G' orgis the CT value of the original Dicom image;

[0097] G' new It is the grayscale value after the window width and window position are adjusted.

[0098] Example 4

[0099] Preferably, the superpixel block in step 3 refers to the average size of all superpixel blocks, and the evaluation method is as follows: assuming that the number of superpixels is n, the size of the original Dicom image is width w, height h, then the average number of superpixel block pixels P is:

[0100]

[0101] Each superpixel is evaluated as a circle-like object, and the average radius of the superpixel is:

[0102]

[0103] As the number of superpixels increases, the average superpixel radius becomes smaller and the degree of refinement becomes higher.

[0104] Preferably, the superpixel calculation in step 7 adopts SEEDS or LSC superpixel algorithm.

[0105] Preferably, the superpixel fine-tuning in step 8 is achieved by adjusting the number of superpixels. By controlling the parameters of the superpixel block size, different numbers of superpixels are selected for target tissues of different scales or different texture complexities to control the segmentation fineness and edge matching, and the CT image is controlled within the range of [500,6000].

[0106] Example 5

[0107] Preferably, the step 10 of performing the intersection operation between the manually marked inner contour and the super pixel block is as follows: recording each pixel point of the manually marked inner contour line and setting it as a set U label , generate the contour binary image F label , where the contour value is 1 and the background is 0; similarly, each superpixel block generates a binary image The method to determine whether the superpixel and the contour intersect is as follows: By traversing all superpixel blocks, all intersection relationships with the contour line are obtained, and the intersecting superpixel blocks are merged to obtain a superpixel block ring, whose outer boundary is the target contour.

[0108] Example 6

[0109] like Figure 1 As shown, a method for quickly marking medical image data using the method of the present invention is as follows:

[0110] Step 1: Figure 2As shown, read in Dicom medical images ( Figure 2 (A original image);

[0111] Step 2: If Figures 2 to 4 As shown, grayscale mapping is performed on Dicom medical images ( Figure 2 After grayscale mapping, there are three cases. If you directly perform superpixel calculation, jump to step 6. If you manually draw a rectangular frame ( Figure 3 If you manually draw the inner contour of the target ( Figure 4 A (rough outline of right kidney), B (rough outline of left kidney), C (vertebral body outline) (red outline), you can skip to step 3;

[0112] Step 3: Manually mark the approximate inner contour of the target, offsetting the true contour by the average radius of one superpixel block, but it must be ensured to be within the target area;

[0113] Step 4: According to the inner contour manually marked in step 3, the maximum and minimum values ​​of the horizontal and vertical coordinates of the inner contour are obtained, and the minimum circumscribed rectangle is determined based on this value, and the size of the 3 super pixel blocks is expanded on this basis to obtain the outer frame ( Figure 4 Medium green outline), skip to step 7;

[0114] Step 5: Manually mark the outer rectangular frame containing the target, define the rectangular frame according to the target size, and jump to step 7;

[0115] Step 6: No marking, directly use the entire Dicom medical image as the outer frame;

[0116] Step 7: Extract the image sub-image corresponding to the outer frame obtained in steps 4, 5, and 6 (such as Figure 5 As shown, A right kidney, B left kidney, C vertebra, D right kidney, E left kidney, F vertebra), calculate superpixels, and obtain multiple superpixel blocks as shown in Figure 6 , Figure 7 As shown, Figure 6 is the result of superpixel calculation of the whole image, A is the superpixel block of the original image, and B is the superpixel block of the windowed image; Figure 7 These are the ROI superpixel calculation results: A is the right kidney, B is the left kidney, C is the vertebra, D is the right kidney, E is the left kidney, and F is the vertebra.

[0117] Step 8: Superpixel fine-tuning. This adjustment can be achieved through two methods to optimize the segmentation contour matching: one is to adjust the window width and window position, and the other is to adjust the number of superpixels, so as to achieve a better target area segmentation effect.

[0118] Step 9: Based on the superpixel calculation results, if the outer contour is manually marked in the previous step, go to step 10, otherwise go to step 11;

[0119] Step 10: Perform an intersection operation on the manually marked inner contour and the superpixel block, and calibrate the superpixel blocks that the contour passes through. The annular outer boundary surrounded by these calibrated superpixel blocks is the target contour (e.g. Figure 8 As shown in the figure, it is a super pixel block diagram of the ROI edge), and all super pixel blocks within the contour are set as the target area through the hole filling algorithm to obtain the marking result (as shown in the figure). Fig. 9 As shown, it is the ROI target outline map);

[0120] Step 11: Manually select superpixel blocks or delineate superpixel blocks in the contour area and automatically fill them, or select and correct the deviated superpixel blocks in step 10, and finally obtain the effective target area and convert it into a binary image. The inner and outer contours of this binary image are mapped to the original image to obtain the labeling result (such as Fig.10 As shown in FIG. 1 , it is the mapping of the target contour to the original image, A is the fusion of the original image contour, and B is the fusion of the window adjustment image contour).

[0121] The principle of the present invention is as follows:

[0122] The algorithm idea of ​​the present invention is that the target marked in the medical image must have a contour that can be discerned by the naked eye. The contour may be clear or blurred. All possible contours are quickly calculated with the assistance of superpixel algorithm, and then superpixel blocks of effective contours are selected manually in a simple way, and finally a binary image of the target area is obtained. This method uses a semi-automatic method to reduce the complexity and refinement requirements of manual operation, and avoids the full contour accuracy of the fully automatic method. At the same time, it has a certain flexibility and can adapt to various target area markings. It solves the problem of manual operation complexity of medical image target contour marking, solves the problem of efficiency of medical image target contour marking, and solves the problem that different medical image scanning parameters affect the versatility of the algorithm.

[0123] The present invention discloses a method for rapid labeling of medical image data based on superpixels, realizes a semi-automatic and universal labeling scheme, and solves many problems in the labeling of medical image organs and tissues, such as quality consistency, target universality, manual workload, program calculation efficiency and calculation amount.

[0124] The present invention cuts the image into blocks through dynamic superpixel calculation, obtains the effective target area by matching the manually marked inner contour with the calculated superpixel block, increases or decreases the superpixel blocks by manual clicking or mouse sliding trajectory, matches the target contour, reduces the amount of calculation, and improves the marking speed.

[0125] The present invention manually marks an outer rectangular frame containing the target, which can be large or small, can reduce the amount of superpixel calculation and the influence of peripheral complex structures and grayscale on superpixel calculation of the target area, and improve the marking speed.

[0126] The manual operation part in the present invention only needs to roughly mark the inner contour of the target manually, and there is no requirement for contour accuracy. The marking speed is fast and the randomness is strong, which can ensure the consistency of marking results of different users and improve marking accuracy.

[0127] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments. Various changes can be made within the knowledge of ordinary geologists in this field without departing from the purpose of the present invention.

[0128] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. A fast labeling method for medical image data based on superpixels, characterized in that: The following steps are involved: Step 1: Read in Dicom medical images; Step 2: Grayscale mapping is performed on the Dicom medical image. There are three cases after grayscale mapping. If superpixel calculation is performed directly, jump to step 6. If the rectangular frame is manually marked, jump to step 5. If the rough inner contour of the target is manually marked, jump to step 3. Step 3: Manually mark the approximate inner contour of the target, offsetting the true contour by the average radius of one superpixel block, but it must be ensured to be within the target area; Step 4: According to the inner contour manually marked in step 3, obtain the maximum and minimum values ​​of the horizontal and vertical coordinates of the inner contour, determine the minimum circumscribed rectangle based on this value, and on this basis, expand the size of 3 to 5 super pixel blocks to generate an outer frame, and jump to step 7; Step 5: Manually mark the outer rectangular box containing the target. Mark the rectangular box according to the target size and jump to step 7. Step 6: No marking, directly use the entire Dicom medical image as the outer frame; Step 7: Extract the image sub-image corresponding to the outer frame obtained in steps 4, 5, and 6, calculate the super-pixel, and obtain multiple super-pixel blocks; Step 8: Superpixel fine-tuning: This fine-tuning optimizes the matching degree of the segmentation contour by selecting one of two methods: one is adjusting the window width and window position, and the other is adjusting the number of superpixels; Step 9: Based on the superpixel calculation results, if the inner contour is manually marked in the previous step, go to step 10, otherwise go to step 11; Step 10: Perform an intersection operation on the manually marked inner contour and the superpixel block, calibrate the superpixel blocks that the contour passes through, and the annular outer boundary surrounded by these calibrated superpixel blocks is the target contour. All superpixel blocks in the contour are set as the target area through the hole filling algorithm to obtain the marking result, and then go to step 11; Step 11: Manually select superpixel blocks or delineate superpixel blocks in the contour area and automatically fill them, or select and correct the deviated superpixel blocks in step 10, and finally obtain the effective target area and convert it into a binary image. The inner and outer contours of this binary image are mapped to the original Dicom medical image as the labeling result.

2. The method for rapid labeling of medical image data based on superpixels according to claim 1, characterized in that: The step 2 performs grayscale mapping on the original Dicom medical image according to the default window width and window position set for the examination item, so that the grayscale values ​​of the Dicom images obtained by different devices and different scanning parameters are controlled within a unified standard range.

3. The method for rapid labeling of medical image data based on superpixels according to claim 2, characterized in that: The specific method of grayscale mapping is as follows: first, a standard image is determined from the sample set, and the window width and window position of the Dicom image are obtained according to the Dicom fields WindowCenter and Window Width. ww and g wc , based on the window width and window position, the effective visible grayscale range is: Then, for any Dicom image, the corresponding effective visible grayscale range can be obtained as g' min and g' max ; Let a gray value of the Dicom image be G' org , then the grayscale mapping calculation formula is as follows: Where: G' org is the CT value of the original Dicom image; G' new It is the grayscale value after the window width and window position are adjusted.

4. The method for rapid labeling of medical image data based on superpixels according to claim 1, characterized in that: The superpixel block in step 3 refers to the average size of all superpixel blocks. The evaluation method is as follows: Assume that the number of superpixels is n, the size of the original Dicom image is width w, height h, then the average number of superpixel blocks P is: Each superpixel is evaluated as a circle-like object, and the average radius of the superpixel is: As the number of superpixels increases, the average superpixel radius becomes smaller and the degree of refinement becomes higher.

5. The method for rapid labeling of medical image data based on superpixels according to claim 1, characterized in that: In step 7, the superpixel calculation adopts SEEDS or LSC superpixel algorithm.

6. The method for rapid labeling of medical image data based on superpixels according to claim 1, characterized in that: In step 8, superpixel fine-tuning is achieved by adjusting the number of superpixels. By controlling the parameters of the superpixel block size, different numbers of superpixels are selected for target tissues of different scales or different texture complexities to control the segmentation fineness and edge matching. The CT image is controlled within the range of [500, 6000].

7. The method for rapid labeling of medical image data based on superpixels according to claim 1, characterized in that: The step 10 of performing the intersection operation between the manually marked inner contour and the super pixel block is as follows: each pixel point of the manually marked inner contour line is recorded and set as a set U label , generate the contour binary image F label , where the contour value is 1 and the background is 0; similarly, each superpixel block generates a binary image The method to determine whether the superpixel and the contour intersect is as follows: By traversing all superpixel blocks, all intersection relationships with the contour line are obtained, and the intersecting superpixel blocks are merged to obtain a superpixel block ring, whose outer boundary is the target contour.

Images