Link algorithm-based diagnostic method for pathological slices of inflammatory bowel disease

Abstract

The invention relates to the technical field of disease diagnosis, in particular to an inflammatory bowel disease pathological slice diagnosis method based on a link algorithm, which comprises the following steps: S1, acquiring an inflammatory bowel disease pathological slice; s2, creating a link algorithm; s3, training the link algorithm in the S2 by using the pathological slice of the inflammatory bowel disease obtained in the S1 to obtain an optimal link algorithm; and S4, inputting pathological slices of the inflammatory bowel disease into the optimal link algorithm in the S3, so as to output diagnosis results of the pathological slices of the inflammatory bowel disease, and enabling the diagnosis results of the plurality of pathological slices of the inflammatory bowel disease to form a report to be exported. According to the method, the link algorithm is established, and the microscopic features of a large number of samples are counted, so that a doctor is assisted in finding potential biomarkers and pathological typing rules of the inflammatory bowel disease, objective quantitative indexes can be provided, subjective deviation of manual interpretation can be reduced, and the consistency of ulcerative colitis diagnosis results can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of disease diagnosis, and in particular to a method for diagnosing inflammatory bowel disease pathology films based on a link algorithm. Background Art

[0002] Inflammatory bowel disease (IBD) is a group of chronic, relapsing gastrointestinal inflammatory diseases of unknown etiology, primarily including ulcerative colitis (UC). Ulcerative colitis: Lesions are limited to the mucosa and submucosa, characterized by continuous, diffuse inflammation. Typical manifestations include crypt abscesses, paucity of goblet cells, and superficial ulcers. Rarely, full-thickness intestinal wall involvement or penetrating complications (e.g., fistulas, strictures) may occur.

[0003] Currently, the pathological diagnosis of ulcerative colitis relies on manual microscopic observation, which presents the following challenges: 1. Pathologists lack unified evaluation standards for quantitative indicators such as inflammatory cell density and crypt structural changes, which affects diagnostic consistency; 2. Traditional methods are unable to efficiently extract quantitative data of microscopic features (e.g., number of crypt cysts, depth of neutrophil infiltration), hindering the study of disease mechanisms and efficacy evaluation of ulcerative colitis. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: in order to solve the technical problem that the existing manual microscope observation method will lead to inconsistent diagnostic results, the present invention provides a diagnostic method for inflammatory bowel disease pathology films based on a link algorithm. By improving the diagnostic method, the consistency of the diagnostic results of ulcerative colitis can be improved.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a diagnostic method for inflammatory bowel disease pathology films based on a link algorithm, comprising the following steps: S1. Obtain pathological films of inflammatory bowel disease; S2, create link algorithm; S3, using the inflammatory bowel disease pathological films obtained in S1 to train the link algorithm in S2 to obtain the optimal link algorithm; S4. Input the pathological slice of inflammatory bowel disease into the optimal link algorithm in S3 to output the diagnosis result of the pathological slice of inflammatory bowel disease, and export the diagnosis results of multiple pathological slices of inflammatory bowel disease in the form of a report.

[0006] Therefore, by creating a link algorithm and counting the microscopic features of a large number of samples, doctors can be assisted in discovering potential biomarkers and pathological classification rules of inflammatory bowel disease. This can provide objective quantitative indicators, reduce subjective bias in manual interpretation, and thus improve the consistency of diagnostic results for ulcerative colitis.

[0007] Furthermore, the link algorithm is the EfficientVit model, and the backbone of the model is the vision transformer.

[0008] Furthermore, the step S2 includes the following steps: S2-1. After obtaining cell samples through rtmdet detection, they are classified through the EfficientVit model to filter out samples with cell impurities, and a dimensionally consistent network with 4D feature implementation and 3D multi-head self-attention is added to offset frequent reshaping operations; S2-2, EfficientVit model output and loss design: The loss function uses the CrossEntropyLoss function. Therefore, in S2-1, the purpose of adding a dimensionally consistent network with 4D feature implementation and 3D multi-head self-attention is to address the resource constraints of edge devices, as Vit-based models are typically several times slower than convolutional networks.

[0009] Furthermore, the S2-1 includes the following steps: S2-1-1, the EfficientVit model is composed of several Patch Embeddings and MetaFormers stacked together; S2-1-2, Dimension Consistency: First, the input image will consist of two , Conv with stride 2 is patched, and then, through Extract low-level semantics and process all Module, and then execute reshape at one time ; S2-1-3, Delay-driven lightweight: According to the task selection module, the meta-path MP is defined, which is a collection of a series of modules. or In the S3 and S4 phases, block selection 、 or .

[0010] Furthermore, in S2-1-1, the final output of EfficientVit The calculation formula is: ; in: yes The number of blocks is also the depth of the network; It is a module consisting of an MLP block and TokenMixer; is the original input image, is the batch size, Represents the height and width of the image respectively.

[0011] ; in: Is the flow to the The intermediate features of the MB modules, Talking-Head Attention, which is essentially a The convolution of The stacked MB modules become a Stage. When different Stages are connected, an embedding operation is used to project the embedding dimension and the downsampling tag length.

[0012] Furthermore, in S2-1-2, the calculation formula for patch embbeding is: ; in: It is The number of channels in a Stage, Represents the output of the original input after patchEmbed, achieving channel dimension reduction and spatial downsampling of 1 / 4.

[0013] ; in, I i Indicates the i The output of a Stage, Pool represents the pooling operation, Conv B Indicates the convolution followed by the BN layer. Conv B,G express Conv B Then connect to GeLU activation layer; ; in, Linear represents a linear layer, MHSA represents multi-head self-attention, LN Represents the LayerNorm layer, Linear G Indicates that the linear layer is connected to the GeLU activation layer, Q Indicates a query, KIndicates the key, V Represents a value, b Indicates bias, T Represents a transpose operation.

[0014] Furthermore, in S2-1-3, the metapaths of the first and second stages , the third and fourth stage metapaths The expression is:

[0015] in, Indicates the first i A 4-dimensional MetaFormer module, Indicates the first i A 3D MetaFormer module, Represents the identity path module.

[0016] Furthermore, the S4 includes the following steps: S4-1, segmenting the pathological tissue of the inflammatory bowel disease pathology slide in S1; S4-2, filtering the pathological tissue segmented in S4-1 to remove impurity interference; S4-3, segmenting the pathological tissue after filtering in S4-2 to extract the crypt area; S4-4, analyze the crypt structure in S4-3 and filter out the crypt structures with incomplete edges; S4-5, segmenting the muscularis mucosa to obtain the mucosal layer, muscularis mucosa, and submucosa; S4-6, classify and diagnose the crypt atrophy of the muscularis mucosa in S4-5; S4-7, crypt deduction of the muscularis mucosa in S4-5; S4-8, classify the effective areas extracted in S4-2 into ulcer and erosion categories; S4-9, extracting the granuloma and suspected granuloma areas of the pathological tissue after filtering in S4-2, and distinguishing granuloma from non-granuloma; S4-10. Segment the crypt structure in S4-4 to distinguish inflammatory cells from normal cells. S4-11, subdividing the cryptitis in S4-4 into normal crypts, cryptitis, suspected cryptitis, and lumen interference; S4-12, subdividing the crypt abscess in S4-4 into crypt invagination, crypt abscess, crypt debris, and normal crypt; S4-13: Outline clipping was used to deduct the crypts in S4-5, and interstitial inflammatory cells were extracted through segmentation to distinguish plasma cells, lymphocytes, eosinophils, and neutrophils. This enabled the detection of crypt abscesses, cryptitis, crypt erosions, ulcers, and interstitial inflammatory cells, as well as crypt location and count; crypt morphology analysis; location of the mucosa, muscularis mucosa, and submucosa; and location and count of interstitial inflammatory cells (including neutrophils, plasma cells, eosinophils, and lymphocytes).

[0017] Furthermore, in S4-6, the classification diagnosis results include: no atrophy, mild atrophy, and severe atrophy.

[0018] Furthermore, in S4-8, the ulcer erosion classification results include: normal, ulcer, and erosion.

[0019] Compared with the prior art, the present invention has the following beneficial effects: By creating a link algorithm and counting the microscopic features of a large number of samples, doctors can be assisted in discovering potential biomarkers and pathological classification patterns of inflammatory bowel disease. This can provide objective quantitative indicators, reduce subjective bias in manual interpretation, and thus improve the consistency of diagnostic results for ulcerative colitis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention will be further described below with reference to the accompanying drawings and examples.

[0021] Figure 1 This is a flow chart of the method for diagnosing inflammatory bowel disease pathology slices based on the link algorithm of the present invention; Figure 2 is a flow chart of the algorithm link of the present invention; Figure 3 This is the Efficientvit architecture diagram of the present invention; Figure 4 Module diagram of MetaFormer and MHSA of the present invention; Figure 5 This is a flow chart of step S4 of the present invention; Figure 6 This is a diagram showing the diagnostic effect of inflammatory bowel disease according to the present invention; Figure 7 An index map for effective region segmentation training of the present invention; Figure 8 This is a diagram showing the segmentation effect of the effective area of the present invention; Figure 9 is the index map of the crypt segmentation training of the present invention; Figure 10 This is a diagram showing the segmentation effect of the crypt of the present invention; Figure 11is an index diagram for the musculosum mucosa segmentation training of the present invention; Figure 12 This is a diagram showing the segmentation effect of the muscularis mucosa of the present invention; Figure 13 This is an indicator diagram for the crypt abscess classification training of the present invention; Figure 14 This is a diagram showing the segmentation effect of a crypt abscess according to the present invention; Figure 15 This is an indicator diagram for the cryptitis classification training of the present invention; Figure 16 This is a classification effect diagram of cryptitis of the present invention; Figure 17 This is an index map for crypt inflammation segmentation training of the present invention; Figure 18 This is a rendering of the segmentation effect of crypt inflammation of the present invention; Figure 19 is an index map for interstitial inflammatory cell segmentation training of the present invention; Figure 20 This is a diagram showing the segmentation effect of interstitial inflammatory cells of the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0023] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, features defined as "first" or "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0024] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0025] like Figures 1 to 20 FIG. 1 is a preferred embodiment of the present invention. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm in this embodiment includes the following steps: S1. Obtain pathological films of inflammatory bowel disease; S2, create link algorithm; S3, using the inflammatory bowel disease pathological films obtained in S1 to train the link algorithm in S2 to obtain the optimal link algorithm; S4. Input the pathological film of inflammatory bowel disease into the optimal link algorithm in S3 to output the diagnostic results of the pathological film of inflammatory bowel disease, and export the diagnostic results of multiple pathological films of inflammatory bowel disease in a report. Thus, by creating a link algorithm, the microscopic features of a large number of samples (for example, the widening degree of the submucosal layer and the distribution frequency of granulomas) are counted to assist doctors in discovering potential biomarkers and pathological classification rules of inflammatory bowel disease. It can provide objective quantitative indicators (for example, crypt branching index and pan-mural inflammation score), reduce the subjective bias of manual interpretation, and thus improve the consistency of the diagnostic results of ulcerative colitis. In addition, it can also promote the integration of multi-center research data, lay a data foundation for the future development of clinical auxiliary diagnosis systems, fill the gap in the quantitative research of inflammatory bowel disease pathology, and provide high-precision, high-throughput analysis tools for IBD pathogenesis research, treatment plan optimization, and prognosis evaluation, helping to transform from empirical medicine to data-driven medicine.

[0026] The link algorithm is the EfficientVit model, whose backbone is the vision transformer; The S2 comprises the following steps: S2-1. After obtaining cell samples through rtmdet detection, they are classified through the EfficientVit model to filter out samples with cell impurities, and a dimensionally consistent network with 4D feature implementation and 3D multi-head self-attention is added to offset frequent reshaping operations; S2-2, EfficientVit model output and loss design: the loss function adopts CrossEntropyLoss function; The S2-1 includes the following steps: S2-1-1, the EfficientVit model is composed of several Patch Embeddings and MetaFormers stacked together; S2-1-2, Dimension Consistency: First, the input image will consist of two , Conv with stride 2 is patched, and then, through Extract low-level semantics and process all Module, and then execute reshape at one time ; S2-1-3, Delay-driven lightweight: According to the task selection module, the meta-path MP is defined, which is a collection of a series of modules. or In the S3 and S4 phases, block selection 、 or ; In S2-1-1, the final output of EfficientVit The calculation formula is: ; in: yes The number of blocks is also the depth of the network; It is a module consisting of an MLP block and TokenMixer; is the original input image, is the batch size, Represents the height and width of the image respectively.

[0027] ; in: Is the flow to the The intermediate features of the MB modules, Talking-Head Attention, which is essentially a The convolution of The stacked MB modules become a stage. When different stages are connected, an embedding operation is used to project the embedding dimension and the downsampled tag length. In S2-1-2, the calculation formula for patch embbeding is: ; in: It is The number of channels in a Stage, Represents the output of the original input after patchEmbed, achieving channel dimension reduction and spatial downsampling of 1 / 4.

[0028] ; in, I i Indicates the i The output of a Stage, Pool represents the pooling operation, Conv B Indicates the convolution followed by the BN layer. Conv B,G express Conv B Then connect to GeLU activation layer; ; in, Linear represents a linear layer, MHSA represents multi-head self-attention, LN Represents the LayerNorm layer, Linear G Indicates that the linear layer is connected to the GeLU activation layer, Q Indicates a query, K Indicates the key, V Represents a value, b Indicates bias, T Represents a transpose operation; In S2-1-3, the metapaths of the first and second stages , the third and fourth stage metapaths The expression is:

[0029] in, Indicates the first i A 4-dimensional MetaFormer module, Indicates the first i A 3D MetaFormer module, Represents the identity path module. Therefore, in S2-1, the purpose of adding a dimensionally consistent network with 4D feature implementation and 3D multi-head self-attention is to: Since Vit-based models are typically several times slower than convolutional networks, it also addresses the issue of limited device resources such as edge devices.

[0030] Specifically, in S2-1-2, the essence of dimensional consistency is to divide the MB module into two networks: 4D and 3D parts. Among them, linear projection and attention are performed on 3D, so that global modeling capabilities can be enjoyed without sacrificing the efficiency of multi-head attention. 4D is mainly used in the beginning part of the network output, mainly implemented in the form of Conv-net.

[0031] In this embodiment, S4 includes the following steps: S4-1, segmenting the pathological tissue of the inflammatory bowel disease pathology slide in S1; S4-2, filtering the pathological tissue segmented in S4-1 to remove impurity interference; S4-3, segmenting the pathological tissue after filtering in S4-2 to extract the crypt area; S4-4, analyze the crypt structure in S4-3 and filter out the crypt structures with incomplete edges; S4-5, segmenting the muscularis mucosa to obtain the mucosal layer, muscularis mucosa, and submucosa; S4-6, classify and diagnose the crypt atrophy of the muscularis mucosa in S4-5; S4-7, crypt deduction of the muscularis mucosa in S4-5; S4-8, classify the effective areas extracted in S4-2 into ulcer and erosion categories; S4-9, extracting the granuloma and suspected granuloma areas of the pathological tissue after filtering in S4-2, and distinguishing granuloma from non-granuloma; S4-10. Segment the crypt structure in S4-4 to distinguish inflammatory cells from normal cells. S4-11, subdividing the cryptitis in S4-4 into normal crypts, cryptitis, suspected cryptitis, and lumen interference; S4-12, subdividing the crypt abscess in S4-4 into crypt invagination, crypt abscess, crypt debris, and normal crypt; S4-13, the crypts in S4-5 were subtracted by contour clipping, and interstitial inflammatory cells were extracted by segmentation to distinguish plasma cells, lymphocytes, eosinophils, and neutrophils; In S4-6, the categorical diagnosis results include: no atrophy, mild atrophy, and severe atrophy; In S4-8, ulcer and erosion classification results include: normal, ulcer, and erosion. This allows for the detection of crypt abscesses, cryptitis, crypt erosion, ulcers, and interstitial inflammatory cells, as well as crypt location and count; crypt morphology analysis; location of the mucosa, muscularis mucosa, and submucosa; and location and count of interstitial inflammatory cells (including neutrophils, plasma cells, eosinophils, and lymphocytes).

[0032] Specifically, the pathological tissue positioning and segmentation of digital pathology films of inflammatory bowel disease was achieved, with a total accuracy of 89.39% for segmentation targets; the crypt structure analysis and positioning of the located effective tissue was performed, with a total accuracy of 93.31%; the mucosal muscle, mucosal layer, and submucosal layer of the located effective tissue were analyzed, with a total accuracy of 61.93%; the presence of crypt abscesses was identified in the filtered crypts, with a total classification accuracy of 95%, of which: the accuracy of crypt abscesses reached 94.3%; the identification of filtered crypts was The accuracy of segmentation was 78.07%, including 80.62% for cryptitis (another 17.81% were suspected cryptitis), 91.8% for normal crypts, and 93.33% for lumen interference. The inflammatory cells were segmented and located in the filtered crypts, with an overall accuracy of 53.62%. The inflammatory cells were segmented and located in the stroma of the mucosal layer (excluding the crypt structure), and quantitative counting was achieved, with an overall accuracy of 66.67%.

[0033] In summary, the present invention creates a link algorithm and counts the microscopic features of a large number of samples to assist doctors in discovering potential biomarkers and pathological classification rules of inflammatory bowel disease. It can provide objective quantitative indicators, reduce subjective bias in manual interpretation, and thus improve the consistency of ulcerative colitis diagnostic results.

[0034] The above description is intended to serve as a guide for the preferred embodiments of the present invention. Based on the above description, relevant personnel are fully capable of making various changes and modifications without departing from the technical scope of the present invention. The technical scope of the present invention is not limited to the contents of the specification and must be determined according to the scope of the claims.

Claims

1. A diagnostic method for inflammatory bowel disease pathology slices based on a link algorithm, characterized in that: The following steps are involved: S1. Obtain pathological films of inflammatory bowel disease; S2, create link algorithm; S3, using the inflammatory bowel disease pathological films obtained in S1 to train the link algorithm in S2 to obtain the optimal link algorithm; S4. Input the pathological slice of inflammatory bowel disease into the optimal link algorithm in S3 to output the diagnosis result of the pathological slice of inflammatory bowel disease, and export the diagnosis results of multiple pathological slices of inflammatory bowel disease in the form of a report.

2. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 1, characterized in that: The link algorithm is the EfficientVit model, and the backbone of this model is the vision transformer.

3. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 1, characterized in that: The S2 comprises the following steps: S2-1. After obtaining cell samples through rtmdet detection, they are classified through the EfficientVit model to filter out samples with cell impurities, and a dimensionally consistent network with 4D feature implementation and 3D multi-head self-attention is added to offset frequent reshaping operations; S2-2. EfficientVit model output and loss design: The loss function adopts the CrossEntropyLoss function.

4. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 3, characterized in that: The S2-1 includes the following steps: S2-1-1, the EfficientVit model is composed of several Patch Embeddings and MetaFormers stacked together; S2-1-2, Dimension Consistency: First, the input image will consist of two , Conv with stride 2 is patched, and then, through Extract low-level semantics and process all Module, and then execute reshape at one time ; S2-1-3, Delay-driven lightweight: According to the task selection module, the meta-path MP is defined, which is a collection of a series of modules. or In the S3 and S4 phases, block selection 、 or .

5. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 4, characterized in that: In S2-1-1, the final output of EfficientVit The calculation formula is: ； in: yes The number of blocks is also the depth of the network; It is a module consisting of an MLP block and TokenMixer; is the original input image, is the batch size, Represents the height and width of the image respectively; ； in: Is the flow to the The intermediate features of the MB modules, Talking-Head Attention, which is essentially a The convolution of The stacked MB modules become a Stage. When different Stages are connected, an embedding operation is used to project the embedding dimension and the downsampling tag length.

6. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 4, characterized in that: In S2-1-2, the calculation formula for patch embbeding is: ； in: It is The number of channels in a Stage, Represents the output of the original input after patchEmbed, achieving channel dimension reduction and spatial downsampling of 1 / 4; ； in, I i Indicates the i The output of a Stage, Pool represents the pooling operation, Conv B Indicates the convolution followed by the BN layer. Conv B,G express Conv B Then connect to GeLU activation layer; ； in, Linear represents a linear layer, MHSA represents multi-head self-attention, LN Represents the LayerNorm layer, Linear G Indicates that the linear layer is connected to the GeLU activation layer, Q Indicates a query, K Indicates the key, V Represents a value, b Indicates bias, T Represents a transpose operation.

7. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 4, characterized in that: In S2-1-3, the metapaths of the first and second stages , the third and fourth stage metapaths The expression is: in, Indicates the first i A 4-dimensional MetaFormer module, Indicates the first i A 3D MetaFormer module, Represents the identity path module.

8. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 1, characterized in that: The S4 comprises the following steps: S4-1, segmenting the pathological tissue of the inflammatory bowel disease pathology slide in S1; S4-2, filtering the pathological tissue segmented in S4-1 to remove impurity interference; S4-3, segmenting the pathological tissue after filtering in S4-2 to extract the crypt area; S4-4, analyze the crypt structure in S4-3 and filter out the crypt structures with incomplete edges; S4-5, segmenting the muscularis mucosa to obtain the mucosal layer, muscularis mucosa, and submucosa; S4-6, classify and diagnose the crypt atrophy of the muscularis mucosa in S4-5; S4-7, crypt deduction of the muscularis mucosa in S4-5; S4-8, classify the effective areas extracted in S4-2 into ulcer and erosion categories; S4-9, extracting the granuloma and suspected granuloma areas of the pathological tissue after filtering in S4-2, and distinguishing granuloma from non-granuloma; S4-10. Segment the crypt structure in S4-4 to distinguish inflammatory cells from normal cells. S4-11, subdividing the cryptitis in S4-4 into normal crypts, cryptitis, suspected cryptitis, and lumen interference; S4-12, subdividing the crypt abscess in S4-4 into crypt invagination, crypt abscess, crypt debris, and normal crypt; S4-13. The crypts in S4-5 were deducted by contour clipping, and interstitial inflammatory cells were extracted by segmentation to distinguish plasma cells, lymphocytes, eosinophils, and neutrophils.

9. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 8, characterized in that: In S4-6, the categorical diagnosis results include: no atrophy, mild atrophy, and severe atrophy.

10. The method for diagnosing inflammatory bowel disease pathology slices based on a link algorithm according to claim 8, characterized in that: In S4-8, the ulcer erosion classification results include: normal, ulcer, and erosion.

Images