Tumor intelligent identification method for colorectal cancer dyeing image

Abstract

The invention relates to the technical field of image processing, in particular to an intelligent tumor identification method for a colorectal cancer dyed image, which comprises the following steps: uniformly dividing the colorectal cancer dyed image into a plurality of sub-regions; determining a region optimization strategy of the corresponding sub-region according to the texture representation value; performing outward expansion by taking the sub-region as a center on the sub-region applicable to the first region optimization strategy, and marking the expanded region as a first tumor suspected region; for the sub-regions applicable to the second region optimization strategy, determining that the sub-regions are independently marked as second tumor suspected regions or are combined with adjacent sub-regions to be marked as the second tumor suspected regions according to comparison results of gland edge fracture matching degrees and preset matching degree thresholds; and determining the early warning level of the colorectal cancer according to the tumor coverage characterization value of the colorectal cancer dyeing image. According to the invention, the accuracy of tumor recognition is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a tumor intelligent recognition method for colorectal cancer staining images. BACKGROUND

[0002] Colorectal cancer is a high-incidence malignant tumor of the digestive tract worldwide, and its early diagnosis and disease assessment are crucial to improving patient survival. In clinical practice, morphological analysis of colorectal cancer pathological staining sections is the "gold standard" for diagnosis. Doctors determine whether there is a tumor and the tumor infiltration range by observing the cell arrangement, gland structure, and cell nucleus morphology in the section. However, manual analysis has limitations such as low efficiency, strong subjectivity, and susceptibility to experience levels. In particular, when faced with a large number of sections or small and extremely weak tumor lesions, the misdiagnosis rate and misdiagnosis rate are significantly increased.

[0003] With the development of medical image intelligent technology, tumor intelligent recognition methods based on computer vision have gradually been applied. The existing mainstream methods mostly use a single model to process the entire image. However, due to the high morphological heterogeneity of colorectal cancer, it is difficult for a single model to maintain high accuracy in typical cancer areas, diffuse infiltration areas, and small lesions, leading to missed detection and misjudgment. Secondly, the grid division in image preprocessing can fragment the complete gland structure, and the model lacks the ability to repair this "artificial break", resulting in fragmented recognition results. Finally, existing methods mostly stop at binary segmentation and cannot provide risk grading with clinical guidance.

[0004] Chinese Patent Application Publication No. CN110390678A discloses a method for segmenting different types of tissues in IHC staining images of colorectal cancer, which includes the following steps: obtaining an IHC staining image, drawing an ROI and extracting image blocks, extracting features from the image blocks, training a classifier, and evaluating the performance of the algorithm in segmenting different types of tissues in the cancerous area of colorectal cancer by calculating the overlap between the segmented image and the ROI region of the same type of tissue in the test group. The present application can extract manual features from the H-stained part of the IHC staining image, train a corresponding classifier, and then identify each image block on the H-stained part of the IHC staining image using the classifier to obtain a segmented image of different types of tissues. This is beneficial for further studying the relationship between the incidence of colorectal cancer and the manual features of different types of tissues in the cancerous area, thereby assisting clinicians in diagnosing cancer patients.

[0005] It can be seen that the above technical solution uses a collection of classification results of independent image blocks, and does not consider that colorectal cancer tumors grow in an infiltrative manner, often crossing multiple adjacent image blocks, affecting tumor assessment results, and thus leading to poor tumor recognition accuracy. SUMMARY

[0006] To this end, the present application provides a tumor intelligent recognition method for colorectal cancer staining images to overcome the problem that the prior art does not consider that the colorectal cancer tumor grows in an invasive manner and often crosses multiple adjacent image blocks, affecting the tumor evaluation result, thereby resulting in poor tumor recognition accuracy.

[0007] To achieve the above-mentioned object, the present application provides a tumor intelligent recognition method for colorectal cancer staining images, comprising: obtaining a colorectal cancer staining image, and uniformly dividing the colorectal cancer staining image into a plurality of sub-regions; calculating texture representation values of the sub-regions, and determining a region optimization strategy for each sub-region according to the texture representation values, wherein the region optimization strategy comprises a first region optimization strategy, a second region optimization strategy, and a third region optimization strategy; performing outward expansion centered on the sub-region to which the first region optimization strategy is applicable, and recording the expanded region as a first tumor suspected region; performing gland edge fracture matching degree calculation on the sub-region to which the second region optimization strategy is applicable, and determining whether to record the sub-region as a second tumor suspected region alone or as a second tumor suspected region after merging with adjacent sub-regions according to the comparison result of the gland edge fracture matching degree and a preset matching degree threshold; performing nucleus area proportion calculation on the sub-region to which the third region optimization strategy is applicable, and determining whether to record the sub-region as a third tumor suspected region or to exclude the sub-region from the suspected tumor region according to the comparison result of the nucleus area proportion and a preset area proportion; fusing the first, second, and third tumor suspected regions to generate a tumor probability heat map, calculating a tumor coverage representation value based on the tumor probability heat map, and determining a warning level of colorectal cancer according to the tumor coverage representation value of the colorectal cancer staining image.

[0008] Further, the process of determining the region optimization strategy for each sub-region according to the texture representation value of the sub-region comprises: comparing the texture representation value with a first preset texture threshold and a second preset texture threshold, respectively; if the texture representation value is less than the first preset texture threshold, the third region optimization strategy is adopted; if the texture representation value is greater than or equal to the first preset texture threshold and less than the second preset texture threshold, the second region optimization strategy is adopted; if the texture representation value is greater than or equal to the second preset texture threshold, the first region optimization strategy is adopted.

[0009] Further, the process of obtaining the texture representation value comprises: graying the sub-region, constructing a gray level co-occurrence matrix of the sub-region, and calculating a texture contrast value of the sub-region; A ratio of the texture contrast value to a preset texture contrast is recorded as the texture feature value.

[0010] Further, the first region optimization strategy comprises: constructing a circular region with a geometric center of the sub-region as a center and a minimum circumscribed circle radius of the sub-region as a radius; increasing the radius of the circular region according to a difference between the texture feature value and the second preset texture threshold, taking a boundary of the circular region after the radius is increased as a new segmentation boundary, and recording a region within the boundary as a first tumor suspected region.

[0011] Further, the increase of the radius of the circular region is provided with several radius adjustment modes, and each adjustment mode has a different increase amplitude of the radius of the circular region.

[0012] Further, the second region optimization strategy comprises: calculating an adenoid edge fracture matching degree of the sub-region; determining an optimization strategy of the sub-region according to the adenoid edge fracture matching degree, wherein, if the adenoid edge fracture matching degree is less than a preset matching degree threshold, recording the sub-region as a second tumor suspected region alone; if the adenoid edge fracture matching degree is greater than or equal to the preset matching degree threshold, merging the sub-region with an adjacent sub-region, and recording a merged region as a second tumor suspected region.

[0013] Further, the process of obtaining the adenoid edge fracture matching degree comprises: extracting edge contours of adenoid structures in the sub-region and the adjacent sub-region, respectively; identifying non-continuous points of the adenoid contours on a side of a common adjacent edge of the sub-region and the adjacent sub-region, and recording the non-continuous points as fracture points; counting a number of the fracture points located on one side of the sub-region as a first number, and counting a number of the fracture points located on one side of the adjacent sub-region as a second number; comparing the first number and the second number, and recording a ratio of a smaller value of the first number and the second number to a larger value of the first number and the second number as the adenoid edge fracture matching degree.

[0014] Further, the third region optimization strategy comprises: calculating a cell nucleus area proportion of the sub-region; determining an optimization strategy of the sub-region according to a nucleus area proportion of the sub-region, wherein, if the nucleus area proportion is less than a preset area proportion, the sub-region is excluded from the suspected tumor region; if the nucleus area proportion is greater than or equal to the preset area proportion, the sub-region is recorded as a third tumor suspected region; the nucleus area proportion is a ratio of an area of all nuclei in the sub-region to a total area of the sub-region.

[0015] Further, the process of determining a warning level of colorectal cancer according to a tumor coverage representation value of the colorectal cancer staining image comprises: comparing the tumor coverage representation value with a first preset tumor coverage threshold and a second preset tumor coverage threshold, respectively; if the tumor coverage representation value is less than the first preset tumor coverage threshold, output a low-risk warning; if the tumor coverage representation value is greater than or equal to the first preset tumor coverage threshold and less than the second preset tumor coverage threshold, output a medium-risk warning; if the tumor coverage representation value is greater than or equal to the second preset tumor coverage threshold, output a high-risk warning.

[0016] Further, the process of obtaining the tumor coverage representation value comprises: fusing the first, second and third tumor suspected regions to form a binary mask of the suspected region; assigning a first confidence score to the first tumor suspected region, a second confidence score to the second tumor suspected region, and a third confidence score to the third tumor suspected region; wherein the first confidence score is greater than the second confidence score, and the second confidence score is greater than the third confidence score; using the first, second and third confidence scores as pixel values to generate a confidence distribution map; performing Gaussian filtering on the confidence distribution map to obtain a tumor probability heat map; recording a region in the tumor probability heat map with a pixel value greater than or equal to a preset pixel value as a tumor region, and calculating a pixel area of the tumor region; recording a ratio of the pixel area of the tumor region to a pixel area of the tumor probability heat map as the tumor coverage representation value.

[0017] Compared with the prior art, the present application has the beneficial effects that: the present application intelligently shunts the image area by the texture characteristic value, guides the high suspected, moderate suspected and low suspected areas to three special processing channels of expansion, merging and fine identification respectively, realizes the leap from "one size fits all" to "one policy for one type", sets the gland edge fracture matching degree, is specially used for detecting and repairing the complete tumor form fragmentation problem caused by the initial image grid division, guarantees the integrity of the tumor identification result, finally generates the tumor probability heat map by fusing the multi-level confidence information, calculates the quantifiable tumor coverage characteristic value according to the tumor probability heat map, and finally outputs the clear three-level risk warning, and builds the complete technical chain of preliminary screening, strategy accurate optimization and risk warning, thereby improving the accuracy of tumor identification.

[0018] Further, the present application divides the texture characteristic value into three clear intervals by setting the first and second preset texture thresholds, each interval corresponds to a unique optimization strategy, avoids the subjective deviation of manual selection of the strategy, and according to the sub-area of high texture characteristic value (close to strong tumor), moderate texture characteristic value (close to weak tumor) and low texture characteristic value (close to normal or extremely weak tumor), the corresponding area division optimization strategy is matched respectively, thereby improving the image recognition accuracy.

[0019] Further, the present application takes the geometric center of the sub-area as the center and the minimum circumscribed circle radius as the radius, ensures that the strong tumor core in the sub-area is completely wrapped, and provides a precise benchmark for subsequent expansion, adjusts the expansion amplitude according to the difference between the texture characteristic value and the second preset texture threshold, accurately matches the pathological characteristics of the strong tumor core of the colorectal cancer to the surrounding mild infiltration, avoids the problem of over-expansion or insufficient expansion caused by fixed radius expansion, completely restores the real infiltration boundary of the tumor, and thereby improves the intelligent level of tumor identification.

[0020] Further, the present application sets several adjustment modes for the radius increase of the circular area, dynamically selects different radius increase amplitudes according to the difference between the texture characteristic value and the threshold, avoids the problem of over-expansion or insufficient expansion caused by a single mode in some scenes, and thereby realizes fine control of the expansion degree of the circular area. Further, the present application identifies the sub-area of the same continuous tumor that is fragmented by the gland edge fracture matching degree, labels the combined sub-area as the second tumor suspected area, solves the defect that the continuous tumor is split into isolated blocks by the rectangular division sub-area, resulting in missed tumor real range, and thereby improves the reliability of tumor identification. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 The flowchart of the tumor intelligent identification method for the colorectal cancer staining image of the embodiment of the present application; Figure 2 The flowchart of determining the area optimization strategy of the corresponding sub-area according to the texture characteristic value of the sub-area of the embodiment of the present application; Figure 3 FIG. 1 shows a flowchart of an embodiment of the present application for determining an optimization strategy for the sub-region according to the gland edge fracture matching degree; Figure 4 FIG. 4 shows a flowchart of an embodiment of the present application for determining a warning level of colorectal cancer according to the tumor coverage characteristic value of the colorectal cancer staining image. DETAILED DESCRIPTION

[0022] In order to make the objects and advantages of the present application clearer, the present application will be further described below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not to limit the present application.

[0023] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present application and not to limit the protection scope of the present application.

[0024] It should be noted that the data in the present embodiment are obtained by comprehensive analysis and evaluation of historical detection data and corresponding historical detection results of the present application in the past three months before the present detection. Those skilled in the art can understand that the determination method of the present application for a single parameter can be to select the value with the highest proportion as the preset standard parameter according to the data distribution, to use weighted summation to obtain the value as the preset standard parameter, to substitute each historical data into a specific formula and to obtain the value as the preset standard parameter using the formula, or other selection methods, as long as the present application can clearly define different specific situations in the single determination process through the obtained value.

[0025] Please refer to FIGS. 1, 2, 3 and 4, which respectively show a flowchart of a tumor intelligent recognition method for a colorectal cancer staining image according to an embodiment of the present application; a flowchart of determining a region optimization strategy for a corresponding sub-region according to a texture characteristic value of the sub-region according to an embodiment of the present application; a flowchart of determining an optimization strategy for the sub-region according to a gland edge fracture matching degree according to an embodiment of the present application; and a flowchart of determining a warning level of colorectal cancer according to a tumor coverage characteristic value of the colorectal cancer staining image according to an embodiment of the present application. Figure 1 Figure 2 Figure 3 Figure 4

[0026] An embodiment of the present application provides a tumor intelligent recognition method for a colorectal cancer staining image, which comprises the following steps. In step S1, a colorectal cancer staining image is obtained, and the colorectal cancer staining image is uniformly divided into a plurality of sub-regions by OpenCV, wherein the 2048*2048 pixel colorectal cancer staining image is uniformly divided into 256*256 pixel sub-regions. ​​​​

[0027] Step S2, calculate the texture characteristic value of each sub-region, and determine the region optimization strategy of the corresponding sub-region according to the texture characteristic value, wherein the region optimization strategy includes a first region optimization strategy, a second region optimization strategy, and a third region optimization strategy; Step S3, performing outward expansion centered on the sub-region to which the first region optimization strategy is applied, and recording the expanded region as a first tumor suspected region; Step S4, performing gland edge fracture matching degree calculation on the sub-region to which the second region optimization strategy is applied, and determining whether to record the sub-region as a second tumor suspected region alone or to record the sub-region as a second tumor suspected region after merging with adjacent sub-regions according to the comparison result of the gland edge fracture matching degree and a preset matching degree threshold; Step S5, performing nucleus area proportion calculation on the sub-region to which the third region optimization strategy is applied, and determining whether to record the sub-region as a third tumor suspected region or to eliminate the sub-region from the suspected tumor region according to the comparison result of the nucleus area proportion and a preset area proportion; Step S6, fusing the first, second, and third tumor suspected regions to generate a tumor probability heat map, calculating a tumor coverage characteristic value based on the tumor probability heat map, and determining a colorectal cancer early warning level according to the tumor coverage characteristic value of the colorectal cancer staining image.

[0028] It should be noted that the data in the present embodiment are all results obtained through preliminary experiments before the present detection by the method of the present application. The preset values can be adjusted according to specific use conditions, as long as the method of the present application can clearly define different specific conditions in the single determination process through the obtained numerical values. The preset values set in the present embodiment are all obtained according to preliminary experiments, and the correction coefficients are also selected through experimental verification.

[0029] Specifically, the process of determining the region optimization strategy of the corresponding sub-region according to the texture characteristic value of the sub-region includes: comparing the texture characteristic value with a first preset texture threshold 1.1 and a second preset texture threshold 1.6, respectively; if the texture characteristic value is less than the first preset texture threshold, the third region optimization strategy is adopted; if the texture characteristic value is greater than or equal to the first preset texture threshold and less than the second preset texture threshold, the second region optimization strategy is adopted; if the texture characteristic value is greater than or equal to the second preset texture threshold, the first region optimization strategy is adopted.

[0030] Specifically, the first preset texture threshold has a value range of [0.8, 1.2], and the second preset texture threshold has a value range of [1.4, 1.8]. Preferably, the first preset texture threshold has a value of 1.1, and the second preset texture threshold has a value of 1.6.

[0031] Specifically, the first preset texture threshold and the second preset texture threshold are key criteria for three-level division of the sub-region texture complexity. In this embodiment, the specific values are obtained by statistically analyzing the distribution of texture characteristic values of 50 normal colorectal mucosa and 50 extremely weak tumor regions in 100 clinical samples.

[0032] Specifically, the process of obtaining the texture characteristic value includes: Based on digital image processing technology, the sub-region is subjected to grayscale processing by programming to call a standard image processing library, a gray level co-occurrence matrix of the sub-region is constructed, and a texture contrast value of the sub-region is calculated. The gray level co-occurrence matrix is a classical statistical tool for describing texture, and its construction and feature extraction are mature and standardized image processing operations. For example, built-in functions for directly calculating the gray level co-occurrence matrix and its derived features are provided in mainstream scientific computing and image processing libraries such as MATLAB, Python's scikit-image, or OpenCV. The step described in the present application can be performed by using any of the above tools or other image processing software / libraries with equivalent functions, and its specific implementation is common knowledge in the art. The ratio of the texture contrast value to the preset texture contrast 110 is denoted as the texture characteristic value.

[0033] The preset texture contrast in this embodiment is selected as 110, and the value is based on the average value of the texture contrasts of all sub-regions calculated from 50 normal colorectal tissue images divided into normal sub-regions, which is denoted as the preset texture contrast. However, the above value is not limited thereto, and those skilled in the art can adjust the above value according to actual needs.

[0034] Specifically, the first region optimization strategy includes: A circular region is constructed with the geometric center of the sub-region as the center and the minimum circumscribed circle radius of the sub-region as the radius. The radius of the circular region is increased according to the difference between the texture characteristic value and the second preset texture threshold, the boundary of the circular region after the radius is increased is used as a new segmentation boundary, and the region within this boundary is denoted as a first tumor suspected region.

[0035] Specifically, the increase in the radius of the circular region is provided with several radius adjustment modes, wherein, If the texture representation difference value is less than a first preset texture representation difference value 0.25, a first adjustment coefficient 1.04 is used to increase the radius of the circular region to a corresponding value; If the texture representation difference value is greater than or equal to the first preset texture representation difference value and less than a second preset texture representation difference value 0.46, a second adjustment coefficient 1.06 is used to increase the radius of the circular region to a corresponding value; If the texture representation difference value is greater than or equal to the second preset texture representation difference value, a third adjustment coefficient 1.08 is used to increase the radius of the circular region to a corresponding value; The texture representation difference value is the difference between the texture representation value and the second preset texture threshold value.

[0036] Specifically, the second region optimization strategy includes: calculating the breast edge fracture matching degree of the sub-region; determining the optimization strategy of the sub-region according to the breast edge fracture matching degree, wherein, If the breast edge fracture matching degree is less than a preset matching degree threshold value 0.75, the sub-region is separately recorded as a second tumor suspected region; If the breast edge fracture matching degree is greater than or equal to the preset matching degree threshold value, the sub-region is merged with the adjacent sub-region, and the merged region is recorded as a second tumor suspected region.

[0037] In this embodiment, 60 groups of clinical samples of sub-regions and adjacent sub-regions meeting the second region optimization strategy are selected, of which 30 groups of the same continuous tumor whose adjacent sub-regions are confirmed by pathologists to be the same tumor and are divided and cracked by a rectangular division, and 30 groups of independent clinical samples whose adjacent sub-regions are confirmed by pathologists to be independent of continuous tumor association; after calculating the breast edge fracture matching degrees of all samples, it is found that the breast edge fracture matching degrees in the same tumor group are concentrated in 0.75-1.00, and the breast edge fracture matching degrees in the independent region group are all less than 0.75, so the preset matching degree threshold value is selected as 0.75, but the above-mentioned values are not limited thereto, and those skilled in the art can adjust the above-mentioned values according to actual needs.

[0038] Specifically, the breast edge fracture matching degree acquisition process includes: The edge contours of the breast structures in the sub-region and the adjacent sub-region are extracted by calling the edge detection algorithm (such as the Canny operator) in the standard image processing library (such as OpenCV); identify a non-continuous point of the gland profile on the side of the common adjacent edge of the sub-region and the adjacent sub-region, and mark the non-continuous point as a broken point, wherein, all adjacent common edge points of the gland profile in the sub-region are traversed, if a certain pixel distance (such as 5 pixels) in the normal direction of a point is not found in the adjacent sub-region, the point is identified as a broken point; the broken point represents the profile position of the same continuous gland structure which is artificially cut off when the sub-region is segmented; count the number of broken points located on one side of the sub-region as a first number, and count the number of broken points located on one side of the adjacent sub-region as a second number; compare the first number and the second number, and mark the ratio of the smaller value in the first number and the second number to the larger value in the first number and the second number as the gland edge broken point matching degree.

[0039] Specifically, the third region optimization strategy includes: calculate the nuclear area proportion of the sub-region; determine the optimization strategy of the sub-region according to the nuclear area proportion of the sub-region, wherein, if the nuclear area proportion is less than a preset area proportion 0.25, the sub-region is excluded from the suspected tumor region; if the nuclear area proportion is greater than or equal to the preset area proportion, the sub-region is marked as a third tumor suspected region; the nuclear area proportion is the ratio of the area of all nuclei in the sub-region to the total area of the sub-region.

[0040] In this embodiment, the nuclear area proportions of 50 groups of normal colorectal mucosa sub-regions are counted, wherein, the nuclei in the normal colorectal mucosa sub-regions are sparse and small in size, and the nuclear area proportions of more than 95% of the 50 groups of normal colorectal mucosa sub-regions are less than 0.25, so the preset area proportion is taken as 0.25, but the above value is not limited thereto, and those skilled in the art can adjust the above value according to actual needs.

[0041] Specifically, the process of determining the early warning level of colorectal cancer according to the tumor coverage representation value of the colorectal cancer staining image includes: compare the tumor coverage representation value with a first preset tumor coverage threshold 0.08 and a second preset tumor coverage threshold 0.35, respectively; if the tumor coverage representation value is less than the first preset tumor coverage threshold, output a low-risk warning; if the tumor coverage representation value is greater than or equal to the first preset tumor coverage threshold and less than the second preset tumor coverage threshold, output a medium-risk warning; If the tumor coverage representation value is greater than or equal to the second preset tumor coverage threshold, a high-risk early warning is output.

[0042] In this embodiment, the first preset tumor coverage threshold has a value range of [0.05, 0.11], and the second preset tumor coverage threshold has a value range of [0.21, 0.42]. Preferably, the first preset tumor coverage threshold has a value of 0.08, and the second preset tumor coverage threshold has a value of 0.35.

[0043] Specifically, the tumor coverage representation value quantifies the proportion of tumor infiltration in the effective tissue area. The greater the tumor coverage representation value, the wider the tumor infiltration range and the more severe the disease.

[0044] Specifically, the tumor coverage representation value is obtained by: The first, second, and third tumor suspected areas are fused to form a binary mask of the suspected area; The first confidence score 1.0 is assigned to the first tumor suspected area, the second confidence score 0.7 is assigned to the second tumor suspected area, and the third confidence score 0.2 is assigned to the third tumor suspected area. The first confidence score is greater than the second confidence score, and the second confidence score is greater than the third confidence score. The first confidence score, the second confidence score, and the third confidence score are used as pixel values to generate a confidence distribution map. In the binary mask, the values of all pixel points located in the first tumor suspected area are set to the first confidence score, the values of all pixel points located in the second tumor suspected area are set to the second confidence score, and the values of all pixel points located in the third tumor suspected area are set to the third confidence score. The pixel value of the non-suspected area is set to 0. Through the above assignment, the generated image is the confidence distribution map.

[0045] The confidence distribution map is subjected to Gaussian filtering to obtain a tumor probability heat map. The area in the tumor probability heat map where the pixel value is greater than or equal to the preset pixel value 0.3 is recorded as a tumor area, and the pixel area of the tumor area is calculated. The ratio of the pixel area of the tumor area to the pixel area of the tumor probability heat map is recorded as the tumor coverage representation value.

[0046] The technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, those skilled in the art will readily understand that the scope of protection of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to the related technical features without departing from the principles of the present application, and the technical solutions after such changes or replacements will all fall within the scope of protection of the present application.

Claims

1. A tumor intelligent recognition method for colorectal cancer staining images, characterized in that, The method comprises the following steps: acquiring a colorectal cancer staining image, and dividing the colorectal cancer staining image into a plurality of sub-regions; calculating texture characteristic values of the sub-regions, and determining a region optimization strategy for each sub-region according to the texture characteristic values, wherein the region optimization strategy comprises a first region optimization strategy, a second region optimization strategy, and a third region optimization strategy; performing outward expansion centered on a sub-region to which the first region optimization strategy is applicable, and recording the expanded region as a first tumor suspected region; performing gland edge fracture matching degree calculation on a sub-region to which the second region optimization strategy is applicable, and determining whether to record the sub-region as a second tumor suspected region alone or as a second tumor suspected region after merging with adjacent sub-regions according to a comparison result of the gland edge fracture matching degree and a preset matching degree threshold; performing nucleus area proportion calculation on a sub-region to which the third region optimization strategy is applicable, and determining whether to record the sub-region as a third tumor suspected region or to exclude the sub-region from the suspected tumor region according to a comparison result of the nucleus area proportion and a preset area proportion; fusing the first, second, and third tumor suspected regions to generate a tumor probability heat map, calculating a tumor coverage characteristic value based on the tumor probability heat map, and determining a warning level of the colorectal cancer according to the tumor coverage characteristic value of the colorectal cancer staining image. 2.The method for tumor intelligent identification of colorectal cancer staining images according to claim 1, characterized in that, The process of determining the region optimization strategy for each sub-region according to the texture characteristic values of the sub-regions comprises the following steps: comparing the texture characteristic values with a first preset texture threshold and a second preset texture threshold, respectively; if the texture characteristic value is less than the first preset texture threshold, the third region optimization strategy is adopted; if the texture characteristic value is greater than or equal to the first preset texture threshold and less than the second preset texture threshold, the second region optimization strategy is adopted; if the texture characteristic value is greater than or equal to the second preset texture threshold, the first region optimization strategy is adopted. 3.The method for tumor intelligent identification of colorectal cancer staining images according to claim 2, characterized in that, The process of obtaining the texture characteristic value comprises the following steps: gray-scale processing the sub-region, constructing a gray-level co-occurrence matrix of the sub-region, and calculating a texture contrast value of the sub-region; the texture contrast value is recorded as the texture characteristic value by comparing the texture contrast value with a preset texture contrast. 4.The method for tumor intelligent identification of colorectal cancer staining images according to claim 3, characterized in that, The first region optimization strategy comprises the following steps: constructing a circular region with the geometric center of the sub-region as the center and the minimum circumscribed circle radius of the sub-region as the radius; increasing the radius of the circular region according to the difference between the texture characteristic value and the second preset texture threshold, taking the boundary of the circular region with the increased radius as a new segmentation boundary, and recording the region within the boundary as a first tumor suspected region. 5.The method for tumor intelligent identification of colorectal cancer staining images according to claim 4, characterized in that, There are several radius adjustment methods for the increase of the radius of the circular region, and each adjustment method has a different increase amplitude for the radius of the circular region. 6.The method for tumor intelligent identification of colorectal cancer staining images according to claim 5, characterized in that, The second region optimization strategy comprises the following steps: calculating the gland edge fracture matching degree of the sub-region; determining the optimization strategy for the sub-region according to the gland edge fracture matching degree, wherein if the gland edge fracture matching degree is less than a preset matching degree threshold, the sub-region is recorded as a second tumor suspected region alone; If the gland edge fracture matching degree is greater than or equal to a preset matching degree threshold, the sub-region and the adjacent sub-region are merged, and the merged region is recorded as a second tumor suspected region. 7.The method for tumor intelligent identification of colorectal cancer staining images according to claim 6, characterized in that, The acquisition process of the gland edge fracture matching degree comprises: The edge contours of the gland structures in the sub-region and the adjacent sub-region are extracted respectively; Non-continuous points of the gland contour are identified on the side of the common adjacent edge of the sub-region and the adjacent sub-region, and the non-continuous points are recorded as fracture points; The number of fracture points located on one side of the sub-region is counted and recorded as a first number; the number of fracture points located on one side of the adjacent sub-region is counted and recorded as a second number; The first number and the second number are compared, and the ratio of the smaller value in the first number and the second number to the larger value in the first number and the second number is recorded as the gland edge fracture matching degree. 8.The method for tumor intelligent identification of colorectal cancer staining images according to claim 7, characterized in that, The third region optimization strategy comprises: The nuclear area proportion of the sub-region is calculated; The optimization strategy of the sub-region is determined according to the nuclear area proportion of the sub-region, wherein, If the nuclear area proportion is less than a preset area proportion, the sub-region is excluded from the suspected tumor region; If the nuclear area proportion is greater than or equal to the preset area proportion, the sub-region is recorded as a third tumor suspected region; The nuclear area proportion is the ratio of the area of all nuclei in the sub-region to the total area of the sub-region. 9.The method for tumor intelligent identification of colorectal cancer staining images according to claim 8, characterized in that, The process of determining the early warning level of colorectal cancer according to the tumor coverage characteristic value of the colorectal cancer staining image comprises: The tumor coverage characteristic value is compared with a first preset tumor coverage threshold and a second preset tumor coverage threshold respectively; If the tumor coverage characteristic value is less than the first preset tumor coverage threshold, a low-risk warning is output; If the tumor coverage characteristic value is greater than or equal to the first preset tumor coverage threshold and less than the second preset tumor coverage threshold, a medium-risk warning is output; If the tumor coverage characteristic value is greater than or equal to the second preset tumor coverage threshold, a high-risk warning is output. 10.The method for tumor intelligent identification of colorectal cancer staining images according to claim 9, characterized in that, The acquisition process of the tumor coverage characteristic value comprises: The first, second and third tumor suspected regions are fused to form a binary mask of the suspected region; The first tumor suspected region is assigned a first confidence score, the second tumor suspected region is assigned a second confidence score, and the third tumor suspected region is assigned a third confidence score; wherein the first confidence score is greater than the second confidence score, and the second confidence score is greater than the third confidence score; The first confidence score, the second confidence score and the third confidence score are taken as pixel values to generate a confidence distribution map; The confidence distribution map is subjected to Gaussian filtering to obtain a tumor probability heat map; Regions in the tumor probability heat map with pixel values greater than or equal to a preset pixel value are recorded as tumor regions, and the pixel area of the tumor regions is calculated; The ratio of the pixel area of the tumor regions to the pixel area of the tumor probability heat map is recorded as the tumor coverage characteristic value.

Images