Evaluation method and device of pathological recognition model applied to pathological diagnosis

Abstract

The invention provides an evaluation method and device of a pathological recognition model applied to pathological diagnosis. The evaluation method comprises the following steps: collecting a plurality of pathological section maps and corresponding diagnosis results; preprocessing the plurality of slice images; dividing the plurality of preprocessed slice images to obtain a plurality of slice image groups; inputting the pre-processed slice images contained in each slice image group into a to-be-evaluated pathological recognition model to obtain a recognition result and diagnosis time corresponding to each pre-processed slice image; according to the recognition result corresponding to each preprocessed slice image in each slice image group, the collected corresponding diagnosis result and the corresponding diagnosis time, generating a plurality of evaluation indexes corresponding to each slice image group; and evaluating the performance of the pathology identification model according to the plurality of evaluation indexes corresponding to each slice image group. The technical effects that the pathological recognition model is accurately evaluated, and the diagnosis accuracy and efficiency of the pathological recognition model in the pathological diagnosis process are ensured are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of model evaluation, and in particular to an evaluation method and device for a pathology recognition model applied to pathology diagnosis. Background Art

[0002] Pathological diagnosis is crucial to clinical treatment and is known as the "gold standard" of medical diagnosis. Pathologists are known as "doctors' doctors". Through the examination and analysis of pathological specimens, they can provide accurate diagnosis and prognosis assessment of diseases, which plays an important role in guiding clinical treatment decisions. However, there is currently an extreme shortage of pathologists worldwide, which poses a huge challenge to pathological diagnosis.

[0003] Taking China as an example, there are only about 10,000 registered pathologists, which is far from meeting the actual demand. According to the recommendation of the National Health Commission, at least 100,000 pathologists are needed to meet clinical needs, which means that the current number of pathologists is only one-tenth of the recommended number. Moreover, the training period for pathologists is long, usually taking five to ten years, which makes it more difficult to solve this problem.

[0004] The shortage of pathologists has had a huge impact on pathological diagnosis. In clinical practice, the accuracy and timeliness of pathological diagnosis are crucial to the life and health of patients. If there are not enough pathologists to perform diagnostic work, it will lead to diagnostic delays, errors and uncertainties, which will bring serious consequences to patients. Therefore, there is an urgent need to change the status quo of pathological diagnosis through technical means.

[0005] In recent years, traditional pathology departments around the world have begun to transform into digital and intelligent ones. More and more large hospitals and medical centers have begun to digitize physical slides and complete diagnosis through fully digital pathology sections. The widespread adoption of fully digital pathology sections has laid the foundation for the research and development and large-scale application of intelligent pathology systems. It is foreseeable that in the next three to five years, global pathology diagnosis will undergo tremendous changes, which is called "new infrastructure for tumor diagnosis."

[0006] In the process of research and development and large-scale application of intelligent pathology systems, pathology recognition models, as one of the most important components, play a vital role. In order to ensure the diagnostic accuracy and efficiency of pathology recognition models in the pathology diagnosis process, it is necessary to evaluate the pathology recognition models based on multiple indicators in order to obtain more accurate evaluation results. However, the current technical solutions lack technical means in this regard.

[0007] Therefore, how to evaluate the pathology recognition model based on multiple indicators in order to obtain more accurate evaluation results, and then determine the appropriate pathology recognition model based on the evaluation results, so as to improve the accuracy and efficiency of pathology diagnosis results by using the appropriate pathology recognition model in the pathology diagnosis process is one of the urgent problems to be solved in the field of model evaluation technology. Summary of the invention

[0008] The present invention aims to solve the technical problems in the above-mentioned technologies to at least a certain extent. To this end, the purpose of the present invention is to provide a method and device for evaluating a pathology recognition model applied to pathology diagnosis, by preprocessing and grouping a number of collected pathology slice images, inputting a pathology recognition model to be evaluated and obtaining multiple groups of multiple evaluation indicators, evaluating the pathology recognition model to be evaluated according to the multiple evaluation indicators and obtaining an evaluation result, thereby evaluating the pathology recognition model based on multiple indicators, obtaining a more accurate evaluation result, and then determining a suitable pathology recognition model according to the evaluation result, and improving the accuracy and efficiency of the pathology diagnosis result by using a suitable pathology recognition model in the pathology diagnosis process.

[0009] The present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, comprising:

[0010] Collect several pathological sections and corresponding pathological diagnosis results;

[0011] Performing image preprocessing on a plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images;

[0012] Dividing a plurality of pre-processed pathological slice images to obtain a plurality of slice image groups;

[0013] Inputting several pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated in turn, and obtaining the pathological recognition result and model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group;

[0014] Generate a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result;

[0015] Generate a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group;

[0016] The performance of the pathology recognition model to be evaluated is evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

[0017] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis collects several pathology slice images, including:

[0018] In the pathology slice image library, M pathology slice images with benign diagnosis results are obtained;

[0019] In the pathology slice image library, N pathology slice images with malignant diagnosis results are obtained;

[0020] M pathological slice images with benign diagnosis results and N pathological slice images with malignant diagnosis results are used as the collected several pathological slice images; wherein M:N=1:1.

[0021] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis performs image preprocessing on a plurality of pathology slice images to obtain a plurality of preprocessed pathology slice images, including:

[0022] Performing image filtering processing on a plurality of pathological slice images to obtain a plurality of filtered pathological slice images;

[0023] Performing image noise reduction processing on a plurality of filtered pathological slice images to obtain a plurality of noise-reduced pathological slice images;

[0024] Performing image enhancement operations on a plurality of denoised pathological slice images to obtain a plurality of enhanced pathological slice images;

[0025] Perform artifact removal operations on a number of enhanced pathology slice images to obtain a number of preprocessed pathology slice images.

[0026] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis performs image filtering processing on a plurality of pathology slice images to obtain a plurality of filtered pathology slice images, including:

[0027] Select any pixel point in any pathological slice image as the target pixel point;

[0028] Determine the pixel on the right side of the target pixel as the right pixel; determine the pixel above the target pixel as the upper pixel;

[0029] Get the pixel values ​​of the target pixel, the right pixel and the upper pixel;

[0030] The difference between the pixel value of the right pixel and the pixel value of the target pixel is taken as the horizontal gradient value G of the target pixel. x ; The difference between the pixel value of the upper pixel and the pixel value of the target pixel is taken as the vertical gradient value G of the target pixel y ;

[0031] According to the horizontal gradient value G x and vertical gradient value G y, calculate the gradient value of the target pixel

[0032] Perform the above operation on all the pixels in the selected pathological slice image to obtain the gradient values ​​of all the pixels in the pathological slice image;

[0033] Determine any row of pixels in the selected pathological slice image, select any pixel to be processed in the row of pixels, and determine the weight of the pixel in the row according to the distance between the pixel in the row and the pixel to be processed;

[0034] Multiply the gradient value of the pixel point corresponding to the pixel point in the row by the weight corresponding to the pixel point, and add the obtained products to obtain the weighted gradient corresponding to the pixel point to be processed;

[0035] The noise standard deviation corresponding to the pixel to be processed is obtained by a preset noise standard deviation acquisition algorithm, and the ratio of the weighted gradient to the noise standard deviation is used as the normalization coefficient of the pixel to be processed;

[0036] The product of the normalization coefficient and the weighted gradient is used as the normalized gradient value of the pixel to be processed;

[0037] The product of the normalized gradient value and the pixel value of the pixel to be processed is used as the filtered pixel value of the pixel to be processed, and the above operation is performed on all the pixels in the selected pathological slice image to obtain the filtered pixel values ​​corresponding to all the pixels in the pathological slice image;

[0038] Obtain the difference between the pixel values ​​of all pixels in the selected pathological slice image and the corresponding filtered pixel values;

[0039] Determine a preset filter parameter corresponding to the selected pathological slice image, and multiply the preset filter parameter by the pixel value of each pixel to obtain a pixel difference threshold corresponding to each pixel;

[0040] When the difference between the pixel value of a pixel point and the corresponding filtered pixel value is greater than the pixel difference threshold, the pixel value of the pixel point remains unchanged; when the difference between the pixel value of a pixel point and the corresponding filtered pixel value is less than or equal to the pixel difference threshold, the pixel value of the pixel point is modified to the corresponding filtered pixel value, and the above operation is performed on all pixels in the pathological slice image to obtain a filtered pathological slice image corresponding to the pathological slice image;

[0041] The above operation is performed on several pathological slice images to obtain several filtered pathological slice images.

[0042] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis performs an artifact removal operation on a plurality of enhanced pathology slice images to obtain a plurality of pre-processed pathology slice images, including:

[0043] Select any enhanced pathological slice image as the pathological slice image to be processed, obtain the pixel values ​​of all pixels in the pathological slice image to be processed, and determine the pixel mean value according to the pixel values ​​of all pixels;

[0044] When the pixel value of a pixel is greater than the pixel mean, the pixel is determined to be a pathological pixel, and all pathological pixels constitute a pathological region; when the pixel value of a pixel is less than or equal to the pixel mean, the pixel is determined to be a background pixel, and all background pixels constitute a background region;

[0045] Performing Radon transform on the pathological area and the background area to obtain corresponding first Radon transform area and second Radon transform area;

[0046] The first Radon transform region and the second Radon transform region are compared, and the part of the second Radon transform region that intersects with the first Radon transform region is taken as the artifact region, and the artifact region is removed to obtain a second Radon correction region corresponding to the second Radon transform region; wherein the part of the second Radon correction region after the artifact region is removed is a blank region;

[0047] Taking the lower left corner of the enhanced pathological slice image as the coordinate origin, a rectangular coordinate system is established, any edge pixel point of the blank area is selected and the coordinates of the edge pixel point are determined, and the pixel values ​​and coordinates of the neighboring pixel points in the four directions of the upper left, lower left, upper right and lower right of the pixel point are determined;

[0048] The difference between the horizontal coordinate of the lower right pixel point and the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the first difference and the second difference, and the ratio of the first difference to the second difference is multiplied by the pixel value of the lower left pixel point to obtain a first product; the difference between the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the third difference, and the ratio of the third difference to the second difference is multiplied by the pixel value of the lower right pixel point to obtain a second product, and the first product is added to the second product to obtain a first pixel filling value of the edge pixel point;

[0049] The difference between the abscissa of the upper right pixel and the abscissa of the edge pixel and the abscissa of the upper left pixel is taken as the fourth difference and the fifth difference, and the ratio of the fourth difference to the fifth difference is multiplied by the pixel value of the upper left pixel to obtain a third product; the difference between the abscissa of the edge pixel and the abscissa of the upper left pixel is taken as the sixth difference, and the ratio of the sixth difference to the fifth difference is multiplied by the pixel value of the upper right pixel to obtain a fourth product, and the third product is added to the fourth product to obtain a second pixel filling value of the edge pixel;

[0050] The difference between the ordinate of the upper right pixel point and the ordinate of the edge pixel point and the ordinate of the lower right pixel point is taken as the seventh difference and the eighth difference, and the ratio of the seventh difference to the eighth difference is multiplied by the first pixel filling value to obtain a fifth product; the difference between the ordinate of the edge pixel point and the ordinate of the lower left pixel point is taken as the ninth difference, and the ratio of the ninth difference to the eighth difference is multiplied by the second pixel filling value to obtain a sixth product, and the fifth product is added to the sixth product to obtain the pixel filling value of the edge pixel point;

[0051] The above operations are performed on all pixels in the blank area in order from the edge to the inside to complete the filling of the blank area, and the second Radon filling area corresponding to the second Radon correction area is obtained;

[0052] Performing an inverse Radon transform on the second Radon filled area to obtain a de-artifacted background area after the artifacts are removed;

[0053] The image area composed of the pathological area and the artifact-removed background area is used as the pre-processed pathological slice image corresponding to the pathological slice image to be processed after the artifacts are removed;

[0054] The above operation is performed on a plurality of enhanced pathological slice images to obtain a plurality of pre-processed pathological slice images corresponding to the plurality of enhanced pathological slice images after the artifact removal operation is performed on the plurality of enhanced pathological slice images.

[0055] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis divides a plurality of pre-processed pathology slice images to obtain a plurality of slice image groups, including:

[0056] Obtaining the number of pre-processed pathological slice images and the preset slice image group division parameters;

[0057] Determine the number of slice groups after division according to the number of sheets and the division parameters;

[0058] According to the number M of preprocessed pathology slice images with benign diagnosis results, the number N of preprocessed pathology slice images with malignant diagnosis results and the number of slice image groups, the number M of preprocessed pathology slice images with benign diagnosis results and the number N of preprocessed pathology slice images with malignant diagnosis results contained in each slice image group are determined; wherein M:N=1:1;

[0059] According to the number M of benign pre-processed pathology slice images and the number N of malignant pre-processed pathology slice images contained in each slice image group, a corresponding pre-processed pathology slice image is allocated to each slice image group, and a number of slice image groups are determined according to the allocation results.

[0060] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis generates a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result, including:

[0061] Select any slice image group as the slice image group to be processed;

[0062] The number of pre-processed pathological slice images with benign pathological identification results in the group of slice images to be processed is recorded as G, and the number of pre-processed pathological slice images with malignant pathological identification results is recorded as B;

[0063] Compare the pathological recognition results in the to-be-processed slice image group with the corresponding collected pathological diagnosis results, record the number of pre-processed pathological slice images whose pathological recognition results and the corresponding collected pathological diagnosis results are both benign as G1, and record the number of pre-processed pathological slice images whose pathological recognition results and the corresponding collected pathological diagnosis results are both malignant as B1;

[0064] The number of pre-processed pathological sections whose pathological identification results are malignant but whose corresponding pathological diagnosis results are benign is recorded as G2;

[0065] According to the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all benign, the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all malignant, the number of pre-processed pathology slice images whose pathological identification results are benign in the slice image group to be processed, and the number of pre-processed pathology slice images whose pathological identification results are malignant, the diagnostic accuracy is calculated as the first evaluation indicator corresponding to the slice image group to be processed;

[0066] Based on the number of preprocessed pathology slice images whose pathology recognition results and the corresponding collected pathology diagnosis results are both benign and the number of preprocessed pathology slice images whose pathology recognition results are malignant but the corresponding collected pathology diagnosis results are benign, the diagnostic sensitivity is calculated as the second evaluation indicator corresponding to the group of slice images to be processed.

[0067] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis generates a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathology slice image contained in each slice image group, including:

[0068] Select any slice image group as the slice image group to be processed;

[0069] Determine the average diagnosis time of the group of slice images to be processed according to the model diagnosis time corresponding to each pre-processed pathological slice image included in the group of slice images to be processed;

[0070] The average diagnosis time is used as the third evaluation indicator corresponding to the slice image group to be processed.

[0071] Preferably, the evaluation method of the pathology recognition model applied to pathology diagnosis evaluates the performance of the pathology recognition model to be evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group, including:

[0072] Determine the relationship between the first evaluation indicator and the first preset threshold, the second evaluation indicator and the second preset threshold, and the third evaluation indicator and the third preset threshold;

[0073] When all slice image groups meet the conditions that the first evaluation index is higher than the first preset threshold, the second evaluation index is higher than the second preset threshold, and the third evaluation index is lower than the third preset threshold, it is determined that the performance evaluation of the pathology recognition model to be evaluated is qualified; otherwise, it is determined that the performance evaluation of the pathology recognition model to be evaluated is unqualified.

[0074] The present invention also provides an evaluation device for a pathology recognition model used in pathology diagnosis, comprising:

[0075] An information collection module is used to collect a number of pathological slice images and a number of corresponding pathological diagnosis results;

[0076] An image preprocessing module is used to perform image preprocessing on a plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images;

[0077] An image division module is used to divide a plurality of pre-processed pathological slice images into a plurality of slice image groups;

[0078] A pathological diagnosis module, used to sequentially input a plurality of pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated, and obtain a pathological recognition result and a model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group;

[0079] Indicator generation module for:

[0080] Generate a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result;

[0081] Generate a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group;

[0082] The performance evaluation module is used to evaluate the performance of the pathology recognition model according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

[0083] The present invention provides an evaluation method and device for a pathology recognition model applied to pathology diagnosis, comprising: collecting a plurality of pathology slice images and a plurality of corresponding pathology diagnosis results; performing image preprocessing on the plurality of pathology slice images to obtain a plurality of preprocessed pathology slice images; dividing the plurality of preprocessed pathology slice images to obtain a plurality of slice image groups; sequentially inputting a plurality of preprocessed pathology slice images contained in each slice image group into a pathology recognition model to be evaluated, and obtaining a pathology recognition result and a model diagnosis time corresponding to each preprocessed pathology slice image in each slice image group; and evaluating the pathology recognition result and the model diagnosis time according to the pathology recognition result and the collected pathology diagnosis results of each preprocessed pathology slice image in each slice image group. Pathological diagnosis results, generate the first evaluation index and the second evaluation index corresponding to each slice image group; generate the third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each pre-processed pathological slice image contained in each slice image group; evaluate the performance of the pathological recognition model to be evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group; thereby achieving the technical effect of evaluating the pathological recognition model based on multiple indicators to obtain more accurate evaluation results, and then determining the appropriate pathological recognition model according to the evaluation results, and improving the accuracy and efficiency of the pathological diagnosis results by using the appropriate pathological recognition model in the pathological diagnosis process.

[0084] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0085] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0087] Figure 1 is a flow chart of an evaluation method of a pathology recognition model applied to pathology diagnosis in an embodiment of the present invention;

[0088] Figure 2 A flowchart for generating an optional evaluation indicator in an embodiment of the present invention;

[0089] Figure 3 The block diagram is a device for evaluating a pathology recognition model used in pathology diagnosis according to an embodiment of the present invention. DETAILED DESCRIPTION

[0090] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0091] Reference Figure 1 , an embodiment of the present invention provides an evaluation method of a pathology recognition model applied to pathology diagnosis, comprising:

[0092] Step S1, collecting a number of pathological slice images and a number of corresponding pathological diagnosis results;

[0093] Step S2, performing image preprocessing on a plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images;

[0094] Step S3, dividing a plurality of pre-processed pathological slice images to obtain a plurality of slice image groups;

[0095] Step S4, sequentially inputting a plurality of pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated, and obtaining a pathological recognition result and a model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group;

[0096] Step S5, generating a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result;

[0097] Step S6, generating a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group;

[0098] Step S7, evaluating the performance of the pathology recognition model to be evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

[0099] In this embodiment, several pathological slice images can be obtained from the pathological slice image library of the hospital.

[0100] In this embodiment, the plurality of pathological slice images include benign pathological slice images and malignant pathological slice images.

[0101] In this embodiment, image preprocessing of the plurality of pathological slice images may include performing noise reduction operations on the plurality of pathological slice images, performing enhancement operations on the plurality of pathological slice images, performing normalization operations on the plurality of pathological slice images, performing artifact removal operations on the plurality of pathological slice images, and the like.

[0102] In this embodiment, a specific implementation method of dividing the plurality of pre-processed pathological slice images may be to divide the plurality of pre-processed pathological slice images according to the types and quantities of the pathological slice images.

[0103] In this embodiment, the pathology recognition model to be evaluated can be trained using existing historical pathology slice images and corresponding historical pathology slice image diagnosis results.

[0104] The technical principle and technical effect of the above technical solution are: collecting a number of pathological slice images and a number of corresponding pathological diagnosis results; performing image preprocessing on a number of pathological slice images to obtain a number of preprocessed pathological slice images; dividing a number of preprocessed pathological slice images to obtain a number of slice image groups; sequentially inputting a number of preprocessed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated to obtain the pathological recognition result and model diagnosis time corresponding to each preprocessed pathological slice image in each slice image group; generating a pathological recognition model according to the pathological recognition result corresponding to each preprocessed pathological slice image in each slice image group and the corresponding pathological diagnosis result collected. A first evaluation indicator and a second evaluation indicator are corresponding to each slice image group; a third evaluation indicator is generated for each slice image group according to the model diagnosis time corresponding to each preprocessed pathology slice image contained in each slice image group; the performance of the pathology recognition model to be evaluated is evaluated according to the first evaluation indicator, the second evaluation indicator and the third evaluation indicator corresponding to each slice image group; thereby achieving the technical effect of evaluating the pathology recognition model based on multiple indicators to obtain a more accurate evaluation result, and then determining a suitable pathology recognition model according to the evaluation result, and improving the accuracy and efficiency of the pathology diagnosis result by using a suitable pathology recognition model in the pathology diagnosis process.

[0105] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, collecting a plurality of pathology slice images, including:

[0106] In the pathology slice image library, M pathology slice images with benign diagnosis results are obtained;

[0107] In the pathology slice image library, N pathology slice images with malignant diagnosis results are obtained;

[0108] M pathological slice images with benign diagnosis results and N pathological slice images with malignant diagnosis results are used as the collected several pathological slice images; wherein M:N=1:1.

[0109] In this embodiment, M>50 and N>50.

[0110] The technical principle and technical effect of the above technical scheme are: in the pathology slice image library, M pathology slice images with benign diagnosis results are obtained; in the pathology slice image library, N pathology slice images with malignant diagnosis results are obtained; M pathology slice images with benign diagnosis results and N pathology slice images with malignant diagnosis results are used as a number of collected pathology slice images; wherein, M:N=1:1, by collecting the same number of benign pathology slice images and malignant pathology slice images, the rationality and completeness of the evaluation process of the pathology recognition model to be evaluated are ensured, the accuracy of the evaluation indicators is further ensured, and the correctness and rationality of the model evaluation results are guaranteed.

[0111] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, which performs image preprocessing on a plurality of pathology slice images to obtain a plurality of preprocessed pathology slice images, including:

[0112] Performing image filtering processing on a plurality of pathological slice images to obtain a plurality of filtered pathological slice images;

[0113] Performing image noise reduction processing on a plurality of filtered pathological slice images to obtain a plurality of noise-reduced pathological slice images;

[0114] Performing image enhancement operations on a plurality of denoised pathological slice images to obtain a plurality of enhanced pathological slice images;

[0115] Perform artifact removal operations on a number of enhanced pathology slice images to obtain a number of preprocessed pathology slice images.

[0116] In this embodiment, the specific implementation method of performing image denoising on several filtered pathological slice images to obtain several denoised pathological slice images can be: select any one of the filtered pathological slice images as the pathological slice image to be processed, select any pixel point in the pathological slice image to be processed as the central pixel point, determine a 9×9 area centered on the central pixel point, obtain the grayscale values ​​of all pixels in the area and sort them in order from large to small, determine the grayscale median, and use the grayscale median as the pixel value of the central pixel point, perform the above operation on all pixels in the pathological slice image to be processed to obtain the denoised pathological slice image corresponding to the pathological slice image to be processed, and perform the above operation on all filtered pathological slice images to obtain several denoised pathological slice images.

[0117] In this embodiment, the specific implementation method of performing image enhancement operation on several denoised pathology slice images to obtain several enhanced pathology slice images can be: select any denoised pathology slice image as the slice image to be processed, select any pixel point in the pathology slice image to be processed as the central pixel point, determine a 9×9 area centered on the central pixel point, obtain the pixel values ​​of all pixels in the area, and then obtain the pixel mean of the area, use the area mean as the pixel value of the central pixel point, perform the above operation on all pixels in the pathology slice image to be processed, and obtain the enhanced pathology slice image corresponding to the pathology slice image to be processed, and perform the above operation on all denoised pathology slice images to obtain several enhanced pathology slice images.

[0118] The technical principle and technical effect of the above technical scheme are: performing image filtering processing on a number of pathological slice images to obtain a number of filtered pathological slice images; performing image denoising processing on a number of filtered pathological slice images to obtain a number of denoised pathological slice images; performing image enhancement operations on a number of denoised pathological slice images to obtain a number of enhanced pathological slice images; performing artifact removal operations on a number of enhanced pathological slice images to obtain a number of preprocessed pathological slice images, thereby achieving the technical effect of improving the clarity of the pathological slice images, enhancing the image details of the pathological slice images, and improving the visual browsing effect of the pathological slice images by preprocessing the pathological slice images, thereby improving the accuracy of the pathological diagnosis results.

[0119] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, which performs image filtering processing on a plurality of pathology slice images to obtain a plurality of filtered pathology slice images, including:

[0120] Select any pixel point in any pathological slice image as the target pixel point;

[0121] Determine the pixel on the right side of the target pixel as the right pixel; determine the pixel above the target pixel as the upper pixel;

[0122] Get the pixel values ​​of the target pixel, the right pixel and the upper pixel;

[0123] The difference between the pixel value of the right pixel and the pixel value of the target pixel is taken as the horizontal gradient value G of the target pixel. x ; The difference between the pixel value of the upper pixel and the pixel value of the target pixel is taken as the vertical gradient value G of the target pixel y ;

[0124] According to the horizontal gradient value G x and vertical gradient value G y , calculate the gradient value of the target pixel

[0125] Perform the above operation on all the pixels in the selected pathological slice image to obtain the gradient values ​​of all the pixels in the pathological slice image;

[0126] Determine any row of pixels in the selected pathological slice image, select any pixel to be processed in the row of pixels, and determine the weight of the pixel in the row according to the distance between the pixel in the row and the pixel to be processed;

[0127] Multiply the gradient value of the pixel point corresponding to the pixel point in the row by the weight corresponding to the pixel point, and add the obtained products to obtain the weighted gradient corresponding to the pixel point to be processed;

[0128] The noise standard deviation corresponding to the pixel to be processed is obtained by a preset noise standard deviation acquisition algorithm, and the ratio of the weighted gradient to the noise standard deviation is used as the normalization coefficient of the pixel to be processed;

[0129] The product of the normalization coefficient and the weighted gradient is used as the normalized gradient value of the pixel to be processed;

[0130] The product of the normalized gradient value and the pixel value of the pixel to be processed is used as the filtered pixel value of the pixel to be processed, and the above operation is performed on all the pixels in the selected pathological slice image to obtain the filtered pixel values ​​corresponding to all the pixels in the pathological slice image;

[0131] Obtain the difference between the pixel values ​​of all pixels in the selected pathological slice image and the corresponding filtered pixel values;

[0132] Determine a preset filter parameter corresponding to the selected pathological slice image, and multiply the preset filter parameter by the pixel value of each pixel to obtain a pixel difference threshold corresponding to each pixel;

[0133] When the difference between the pixel value of a pixel point and the corresponding filtered pixel value is greater than the pixel difference threshold, the pixel value of the pixel point remains unchanged; when the difference between the pixel value of a pixel point and the corresponding filtered pixel value is less than or equal to the pixel difference threshold, the pixel value of the pixel point is modified to the corresponding filtered pixel value, and the above operation is performed on all pixels in the pathological slice image to obtain a filtered pathological slice image corresponding to the pathological slice image;

[0134] The above operation is performed on several pathological slice images to obtain several filtered pathological slice images.

[0135] In this embodiment, a specific implementation method of determining any row of pixels in the selected pathological slice image, selecting any pixel to be processed in the row of pixels, and determining the weight of the pixel in the row according to the distance between the pixel in the row and the pixel to be processed can be: setting the weight formula to W = e -x, where W is the weight of a pixel in the row, e is the base of the exponential function, which is 2.7 here, and x is the distance between a pixel in the row and the pixel to be processed, that is, the number of pixels between a pixel in the row and the pixel to be processed. The weight of the pixel in the row is determined according to the weight formula and the distance between the pixel in the row and the pixel to be processed. The farther away from the pixel to be processed, the smaller the weight of the pixel.

[0136] In this embodiment, the gradient value of the pixel point corresponding to the row of pixels is multiplied by the weight corresponding to the pixel point, and the obtained products are added to obtain the weighted gradient corresponding to the pixel point to be processed. The specific implementation method may be: determine that there are N pixels in the row at this time, and the corresponding gradient values ​​of the N pixels are G1, G2, G3, ..., G N , the corresponding weights of N pixels are W1, W2, W3, ..., W N , the weighted gradient corresponding to the pixel to be processed is G1W1+G2W2+G3W3+...+G N W N .

[0137] In this embodiment, the preset noise standard deviation acquisition algorithm can be obtained from a code website, such as the CSDN website.

[0138] In this embodiment, the preset filtering parameter may be any value between 0 and 1.

[0139] The technical principle and technical effect of the above technical scheme are: first, the gradient value of the pixel is obtained according to the pixel values ​​of the pixel point, the pixel point above the pixel point and the pixel point to the right of the pixel point in the pathological slice image, then each row of pixels is processed, the pixel weight is determined according to the pixel distance, the weighted gradient value of the pixel point is determined according to the gradient value and weight of the pixel point, the filtered pixel value of the pixel point is determined according to the pixel value of the pixel point, the weighted gradient value and the noise standard deviation, and then the pixel difference threshold is determined according to the preset filtering parameters and the pixel value of the pixel point, and finally the pixel value of the pixel point is determined according to the difference between the pixel value of the pixel point and the corresponding filtered pixel value and the pixel difference threshold of the pixel point, to complete the filtering of the pathological slice image, thereby improving the clarity of the pathological slice image, enhancing the image detail characteristics of the pathological slice image, improving the user's visual browsing effect of the pathological slice image, and thereby improving the correctness of the pathological diagnosis result.

[0140] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, performing an artifact removal operation on a plurality of enhanced pathology slice images to obtain a plurality of pre-processed pathology slice images, including:

[0141] Select any enhanced pathological slice image as the pathological slice image to be processed, obtain the pixel values ​​of all pixels in the pathological slice image to be processed, and determine the pixel mean value according to the pixel values ​​of all pixels;

[0142] When the pixel value of a pixel is greater than the pixel mean, the pixel is determined to be a pathological pixel, and all pathological pixels constitute a pathological region; when the pixel value of a pixel is less than or equal to the pixel mean, the pixel is determined to be a background pixel, and all background pixels constitute a background region;

[0143] Performing Radon transform on the pathological area and the background area to obtain corresponding first Radon transform area and second Radon transform area;

[0144] The first Radon transform region and the second Radon transform region are compared, and the part of the second Radon transform region that intersects with the first Radon transform region is taken as the artifact region, and the artifact region is removed to obtain a second Radon correction region corresponding to the second Radon transform region; wherein the part of the second Radon correction region after the artifact region is removed is a blank region;

[0145] Taking the lower left corner of the enhanced pathological slice image as the coordinate origin, a rectangular coordinate system is established, any edge pixel point of the blank area is selected and the coordinates of the edge pixel point are determined, and the pixel values ​​and coordinates of the neighboring pixel points in the four directions of the upper left, lower left, upper right and lower right of the pixel point are determined;

[0146] The difference between the horizontal coordinate of the lower right pixel point and the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the first difference and the second difference, and the ratio of the first difference to the second difference is multiplied by the pixel value of the lower left pixel point to obtain a first product; the difference between the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the third difference, and the ratio of the third difference to the second difference is multiplied by the pixel value of the lower right pixel point to obtain a second product, and the first product is added to the second product to obtain a first pixel filling value of the edge pixel point;

[0147] The difference between the abscissa of the upper right pixel and the abscissa of the edge pixel and the abscissa of the upper left pixel is taken as the fourth difference and the fifth difference, and the ratio of the fourth difference to the fifth difference is multiplied by the pixel value of the upper left pixel to obtain a third product; the difference between the abscissa of the edge pixel and the abscissa of the upper left pixel is taken as the sixth difference, and the ratio of the sixth difference to the fifth difference is multiplied by the pixel value of the upper right pixel to obtain a fourth product, and the third product is added to the fourth product to obtain a second pixel filling value of the edge pixel;

[0148] The difference between the ordinate of the upper right pixel point and the ordinate of the edge pixel point and the ordinate of the lower right pixel point is taken as the seventh difference and the eighth difference, and the ratio of the seventh difference to the eighth difference is multiplied by the first pixel filling value to obtain a fifth product; the difference between the ordinate of the edge pixel point and the ordinate of the lower left pixel point is taken as the ninth difference, and the ratio of the ninth difference to the eighth difference is multiplied by the second pixel filling value to obtain a sixth product, and the fifth product is added to the sixth product to obtain the pixel filling value of the edge pixel point;

[0149] The above operations are performed on all pixels in the blank area in order from the edge to the inside to complete the filling of the blank area, and the second Radon filling area corresponding to the second Radon correction area is obtained;

[0150] Performing an inverse Radon transform on the second Radon filled area to obtain a de-artifacted background area after the artifacts are removed;

[0151] The image area composed of the pathological area and the artifact-removed background area is used as the pre-processed pathological slice image corresponding to the pathological slice image to be processed after the artifacts are removed;

[0152] The above operation is performed on a plurality of enhanced pathological slice images to obtain a plurality of pre-processed pathological slice images corresponding to the plurality of enhanced pathological slice images after the artifact removal operation is performed on the plurality of enhanced pathological slice images.

[0153] In this embodiment, the specific implementation codes of Radon transform and inverse Radon transform can be obtained from a code website, such as the CSDN website.

[0154] In this embodiment, a specific implementation method for determining the neighboring pixel points in the upper left, lower left, upper right and lower right directions of the pixel point can be: establish a sub-coordinate system with the pixel point as the origin, draw extension lines with the origin as the vertex in the northwest (with an angle of 45° to the vertical upward direction), southwest (with an angle of 45° to the vertical downward direction), northeast (with an angle of 45° to the vertical upward direction) and southeast (with an angle of 45° to the vertical downward direction), and take the four pixel points on the four extension lines closest to the pixel point as the neighboring pixel points in the upper left, lower left, upper right and lower right directions of the pixel point.

[0155] The technical principle and technical effect of the above technical scheme are: by comparing the pixel value of the pixel point in the pathological slice image with the pixel mean value, the background area and the pathological area are divided; by performing Radon transform on the two areas and comparing the corresponding areas after the transform, the artifact area in the background area is determined and removed; the blank area left after the removal is interpolated and filled; the filled background area is subjected to inverse Radon transform to obtain the artifact-free background area; the image area composed of the pathological area and the artifact-free background area is used as the new image area to complete the image artifact removal operation, thereby improving the clarity of the pathological slice image, enhancing the image detail features of the pathological slice image, avoiding user visual errors caused by artifacts, improving the user's visual browsing effect of the pathological slice image, and thereby improving the correctness of the pathological diagnosis result.

[0156] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, which divides a plurality of pre-processed pathology slice images to obtain a plurality of slice image groups, including:

[0157] Obtaining the number of pre-processed pathological slice images and the preset slice image group division parameters;

[0158] Determine the number of slice groups after division according to the number of sheets and the division parameters;

[0159] According to the number M of preprocessed pathology slice images with benign diagnosis results, the number N of preprocessed pathology slice images with malignant diagnosis results and the number of slice image groups, the number M of preprocessed pathology slice images with benign diagnosis results and the number N of preprocessed pathology slice images with malignant diagnosis results contained in each slice image group are determined; wherein M:N=1:1;

[0160] According to the number M of benign pre-processed pathology slice images and the number N of malignant pre-processed pathology slice images contained in each slice image group, a corresponding pre-processed pathology slice image is allocated to each slice image group, and a number of slice image groups are determined according to the allocation results.

[0161] In this embodiment, the preset slice image group division parameter can be set according to the number of pre-processed pathological slice images. For example, when the number of pre-processed pathological slice images is 200, the preset slice image group division parameter can be 20.

[0162] In this embodiment, the specific implementation method of determining the number of slice image groups after division according to the number of images and the division parameter can be: the number of images is determined to be Q, the preset slice image group division parameter is D, and the number of slice image groups is the value obtained by rounding up the value obtained by Q / D. For example, if Q / D is 10.5, the number of slice image groups is 11, wherein each of the first 10 groups has D pre-processed pathology slice images, and the 11th group has (Q-10D) pre-processed pathology slice images.

[0163] In this embodiment, each slice image group is provided with a group number, and the group numbers may be P1, P2, P3, ..., P n .

[0164] In this embodiment, the number of pre-processed pathological slice images with benign diagnosis results included in each slice image group is determined according to the number M of pre-processed pathological slice images with benign diagnosis results, the number N of pre-processed pathological slice images with malignant diagnosis results, and the number of slice image groups. and the number of pre-processed pathological sections with a malignant diagnosis The specific implementation method may be: when the number of groups is Z, the first Z-1 groups of each group All are the values ​​obtained by rounding down the values ​​obtained by M / Z. The values ​​are all obtained by rounding down the values ​​obtained by N / Z. for for in The value obtained by rounding down the value of M / Z, The value obtained by rounding down the value obtained by dividing N by Z.

[0165] The technical principle and technical effect of the above technical solution are: obtaining the number of pre-processed pathological slice images and the preset slice image group division parameters; determining the number of slice image groups after division according to the number of images and the division parameters; determining the number of pre-processed pathological slice images with benign diagnosis results contained in each slice image group according to the number M of pre-processed pathological slice images with benign diagnosis results, the number N of pre-processed pathological slice images with malignant diagnosis results and the number of slice image groups. and the number of pre-processed pathological sections with a malignant diagnosis in, According to the number of benign pre-processed pathological slices contained in each slice group and the number of malignant pre-processed pathological sections A corresponding preprocessed pathology slice image is assigned to each slice image group, and a number of slice image groups are determined according to the assignment results, thereby achieving the technical effect of ensuring the rationality and completeness of the evaluation process of the pathology recognition model to be evaluated, further ensuring the accuracy of the evaluation indicators, and then ensuring the correctness and rationality of the model evaluation results.

[0166] Reference Figure 2 The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, which generates a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the corresponding collected pathology diagnosis result, including:

[0167] Step S51, selecting any slice image group as the slice image group to be processed;

[0168] Step S52, the number of pre-processed pathological slice images with benign pathological identification results in the slice image group to be processed is recorded as G, and the number of pre-processed pathological slice images with malignant pathological identification results is recorded as B;

[0169] Step S53, comparing the pathological identification results in the to-be-processed slice image group with the corresponding collected pathological diagnosis results, recording the number of pre-processed pathological slice images whose pathological identification results and the corresponding collected pathological diagnosis results are both benign as G1, and recording the number of pre-processed pathological slice images whose pathological identification results and the corresponding collected pathological diagnosis results are both malignant as B1;

[0170] Step S54, the number of pre-processed pathological slice images whose pathological identification results are malignant but whose corresponding pathological diagnosis results are benign is recorded as G2;

[0171] Step S55, according to the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all benign, the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all malignant, the number of pre-processed pathology slice images whose pathological identification results are benign in the slice image group to be processed, and the number of pre-processed pathology slice images whose pathological identification results are malignant, the diagnosis accuracy is calculated as the first evaluation index corresponding to the slice image group to be processed;

[0172] Step S56, based on the number of pre-processed pathology slice images whose pathology recognition results and the collected corresponding pathology diagnosis results are both benign and the number of pre-processed pathology slice images whose pathology recognition results are malignant but the collected corresponding pathology diagnosis results are benign, calculate the diagnostic sensitivity as the second evaluation indicator corresponding to the slice image group to be processed.

[0173] In this embodiment, the calculation formula of the first evaluation index is Right=(G1+B1) / (G+B), where Right is the diagnosis accuracy rate.

[0174] In this embodiment, the calculation formula of the second evaluation index is Sensitivity=G1 / (G1+G2), where Sensitivity is the diagnostic sensitivity.

[0175] The technical principle and technical effect of the above technical scheme are: according to the number of pre-processed pathology slice images with benign pathology recognition results in the slice image group, the number of pre-processed pathology slice images with malignant pathology recognition results, the number of pre-processed pathology slice images with both benign pathology recognition results and the corresponding collected pathology diagnosis results, the number of pre-processed pathology slice images with both malignant pathology recognition results and the corresponding collected pathology diagnosis results, and the number of pathology slice images with malignant pathology recognition results but benign corresponding pathology diagnosis results, the first evaluation index and the second evaluation index corresponding to the slice group are determined, thereby achieving the technical effect of obtaining multiple evaluation indexes of the pathology recognition model to be evaluated, improving the accuracy and rationality of the evaluation results, and determining the diagnostic accuracy of the model.

[0176] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, generating a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathology slice image included in each slice image group, including:

[0177] Select any slice image group as the slice image group to be processed;

[0178] Determine the average diagnosis time of the group of slice images to be processed according to the model diagnosis time corresponding to each pre-processed pathological slice image included in the group of slice images to be processed;

[0179] The average diagnosis time is used as the third evaluation indicator corresponding to the slice image group to be processed.

[0180] The technical principle and technical effect of the above technical solution are: selecting any slice image group as the slice image group to be processed; determining the average diagnosis time of the slice image group to be processed according to the model diagnosis time corresponding to each pre-processed pathological slice image contained in the slice image group to be processed; using the average diagnosis time as the third evaluation indicator corresponding to the slice image group to be processed, realizing the technical effect of using the average pathological diagnosis time as the evaluation indicator, improving the accuracy and rationality of the evaluation result, and determining the model diagnosis speed

[0181] The embodiment of the present invention provides an evaluation method for a pathology recognition model applied to pathology diagnosis, which evaluates the performance of the pathology recognition model to be evaluated according to a first evaluation index, a second evaluation index, and a third evaluation index corresponding to each slice image group, including:

[0182] Determine the relationship between the first evaluation indicator and the first preset threshold, the second evaluation indicator and the second preset threshold, and the third evaluation indicator and the third preset threshold;

[0183] When all slice image groups meet the conditions that the first evaluation index is higher than the first preset threshold, the second evaluation index is higher than the second preset threshold, and the third evaluation index is lower than the third preset threshold, it is determined that the performance evaluation of the pathology recognition model to be evaluated is qualified; otherwise, it is determined that the performance evaluation of the pathology recognition model to be evaluated is unqualified.

[0184] In this embodiment, the first preset threshold is a value between 0.95-0.98, the second preset threshold is a value between 0.9-0.95, and the third preset threshold is 23s.

[0185] The technical principle and technical effect of the above technical scheme are: judging the relationship between the first evaluation indicator and the first preset threshold, the second evaluation indicator and the second preset threshold, and the third evaluation indicator and the third preset threshold; when the conditions that the first evaluation indicator is higher than the first preset threshold, the second evaluation indicator is higher than the second preset threshold, and the third evaluation indicator is lower than the third preset threshold are met, determining that the performance evaluation of the pathology recognition model to be evaluated is qualified; when the three evaluation indicators do not meet the conditions, determining that the performance evaluation of the pathology recognition model to be evaluated is unqualified, realizing the evaluation of the pathology recognition model to be evaluated by multiple indicators, and then obtaining a more accurate evaluation result, thereby determining a suitable pathology recognition model according to the evaluation result, and improving the accuracy and efficiency of the pathology diagnosis result by using a suitable pathology recognition model in the pathology diagnosis process.

[0186] Reference Figure 3 The embodiment of the present invention further provides an evaluation device for a pathology recognition model applied to pathology diagnosis, comprising:

[0187] The information collection module,10,is used to collect a number of pathological slices and a number of corresponding pathological diagnosis results;

[0188] An image preprocessing module 20 is used to perform image preprocessing on a plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images;

[0189] An image division module 30 is used to divide a plurality of pre-processed pathological slice images into a plurality of slice image groups;

[0190] The pathological diagnosis module 40 is used to sequentially input a plurality of pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated, and obtain the pathological recognition result and model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group;

[0191] The indicator generation module 50 is used to:

[0192] Generate a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result;

[0193] Generate a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group;

[0194] The performance evaluation module 60 is used to evaluate the performance of the pathology recognition model according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

[0195] In this embodiment, several pathological slice images can be obtained from the pathological slice image library of the hospital.

[0196] In this embodiment, the plurality of pathological slice images include benign pathological slice images and malignant pathological slice images.

[0197] In this embodiment, image preprocessing of the plurality of pathological slice images may include performing noise reduction operations on the plurality of pathological slice images, performing enhancement operations on the plurality of pathological slice images, performing normalization operations on the plurality of pathological slice images, performing artifact removal operations on the plurality of pathological slice images, and the like.

[0198] In this embodiment, a specific implementation method of dividing the plurality of pre-processed pathological slice images may be to divide the plurality of pre-processed pathological slice images according to the types and quantities of the pathological slice images.

[0199] In this embodiment, the pathology recognition model to be evaluated can be trained using existing historical pathology slice images and corresponding historical pathology slice image diagnosis results.

[0200] The technical principle and technical effect of the above technical solution are: collecting a number of pathological slice images and a number of corresponding pathological diagnosis results; performing image preprocessing on a number of pathological slice images to obtain a number of preprocessed pathological slice images; dividing a number of preprocessed pathological slice images to obtain a number of slice image groups; sequentially inputting a number of preprocessed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated to obtain the pathological recognition result and model diagnosis time corresponding to each preprocessed pathological slice image in each slice image group; generating a pathological recognition model according to the pathological recognition result corresponding to each preprocessed pathological slice image in each slice image group and the corresponding pathological diagnosis result collected. A first evaluation indicator and a second evaluation indicator are corresponding to each slice image group; a third evaluation indicator is generated for each slice image group according to the model diagnosis time corresponding to each preprocessed pathology slice image contained in each slice image group; the performance of the pathology recognition model to be evaluated is evaluated according to the first evaluation indicator, the second evaluation indicator and the third evaluation indicator corresponding to each slice image group; thereby achieving the technical effect of evaluating the pathology recognition model based on multiple indicators to obtain a more accurate evaluation result, and then determining a suitable pathology recognition model according to the evaluation result, and improving the accuracy and efficiency of the pathology diagnosis result by using a suitable pathology recognition model in the pathology diagnosis process.

[0201] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for evaluating a pathology recognition model for pathology diagnosis, characterized in that: include: Collect several pathological sections and corresponding pathological diagnosis results; Performing image preprocessing on the plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images; Dividing the plurality of pre-processed pathological slice images to obtain a plurality of slice image groups; Inputting several pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated in turn, and obtaining the pathological recognition result and model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group; Generate a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result; Generate a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group; The performance of the pathology recognition model to be evaluated is evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

2. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 1, characterized in that: Collect several pathological sections, including: In the pathology slice image library, M pathology slice images with benign diagnosis results are obtained; In the pathology slice image library, N pathology slice images with malignant diagnosis results are obtained; M pathological slice images with benign diagnosis results and N pathological slice images with malignant diagnosis results are used as the collected several pathological slice images; wherein M:N=1:

1.

3. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 1, characterized in that: Performing image preprocessing on the plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images includes: Performing image filtering processing on the plurality of pathological slice images to obtain a plurality of filtered pathological slice images; Performing image noise reduction processing on the plurality of filtered pathological slice images to obtain a plurality of noise-reduced pathological slice images; Performing image enhancement operation on the plurality of denoised pathological slice images to obtain a plurality of enhanced pathological slice images; Performing an artifact removal operation on the plurality of enhanced pathological slice images to obtain a plurality of preprocessed pathological slice images.

4. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 3, characterized in that: Performing image filtering processing on the plurality of pathological slice images to obtain a plurality of filtered pathological slice images, including: Select any pixel point in any pathological slice image as the target pixel point; Determine the pixel on the right side of the target pixel as the right pixel; determine the pixel above the target pixel as the upper pixel; Get the pixel values ​​of the target pixel, the right pixel and the upper pixel; The difference between the pixel value of the right pixel and the pixel value of the target pixel is taken as the horizontal gradient value G of the target pixel. x ; The difference between the pixel value of the upper pixel and the pixel value of the target pixel is taken as the vertical gradient value G of the target pixel y ; According to the horizontal gradient value G x and vertical gradient value G y , calculate the gradient value of the target pixel Perform the above operation on all the pixels in the selected pathological slice image to obtain the gradient values ​​of all the pixels in the pathological slice image; Determine any row of pixels in the selected pathological slice image, select any pixel to be processed in the row of pixels, and determine the weight of the pixel in the row according to the distance between the pixel in the row and the pixel to be processed; Multiplying the gradient value of the pixel point corresponding to the pixel point in the row by the weight corresponding to the pixel point, and adding the obtained products to obtain the weighted gradient corresponding to the pixel point to be processed; The noise standard deviation corresponding to the pixel to be processed is obtained by a preset noise standard deviation acquisition algorithm, and the ratio of the weighted gradient to the noise standard deviation is used as the normalization coefficient of the pixel to be processed; Taking the product of the normalization coefficient and the weighted gradient as the normalized gradient value of the pixel to be processed; The product of the normalized gradient value and the pixel value of the pixel to be processed is used as the filtered pixel value of the pixel to be processed, and the above operation is performed on all the pixels in the selected pathological slice image to obtain the filtered pixel values ​​corresponding to all the pixels in the pathological slice image; Obtaining the difference between the pixel values ​​of all pixels in the selected pathological slice image and the corresponding filtered pixel values; Determine a preset filtering parameter corresponding to the selected pathological slice image, and multiply the preset filtering parameter by the pixel value of each pixel to obtain a pixel difference threshold corresponding to each pixel; When the difference between the pixel value of the pixel point and the corresponding filtered pixel value is greater than the pixel difference threshold, the pixel value of the pixel point remains unchanged; when the difference between the pixel value of the pixel point and the corresponding filtered pixel value is less than or equal to the pixel difference threshold, the pixel value of the pixel point is modified to the corresponding filtered pixel value, and the above operation is performed on all the pixels in the pathological slice image to obtain a filtered pathological slice image corresponding to the pathological slice image; The above operation is performed on several pathological slice images to obtain several filtered pathological slice images.

5. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 3, characterized in that: Performing an artifact removal operation on the plurality of enhanced pathological slice images to obtain a plurality of preprocessed pathological slice images, including: Select any enhanced pathological slice image as the pathological slice image to be processed, obtain the pixel values ​​of all pixels in the pathological slice image to be processed, and determine the pixel mean according to the pixel values ​​of all pixels; When the pixel value of the pixel point is greater than the pixel mean, the pixel point is determined to be a pathological pixel point, and all pathological pixel points constitute a pathological area; when the pixel value of the pixel point is less than or equal to the pixel mean, the pixel point is determined to be a background pixel point, and all background pixel points constitute a background area; Performing Radon transform on the pathological region and the background region to obtain corresponding first Radon transform region and second Radon transform region; The first Radon transform region and the second Radon transform region are compared, and the part of the second Radon transform region that intersects with the first Radon transform region is taken as the artifact region, and the artifact region is removed to obtain a second Radon correction region corresponding to the second Radon transform region; wherein the part of the second Radon correction region after the artifact region is removed is a blank region; Taking the lower left corner of the enhanced pathological slice image as the coordinate origin, a rectangular coordinate system is established, any edge pixel point of the blank area is selected and the coordinates of the edge pixel point are determined, and the pixel values ​​and coordinates of the neighboring pixel points in the upper left, lower left, upper right and lower right directions of the pixel point are determined; The difference between the horizontal coordinate of the lower right pixel point and the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the first difference and the second difference, and the ratio of the first difference to the second difference is multiplied by the pixel value of the lower left pixel point to obtain a first product; the difference between the horizontal coordinate of the edge pixel point and the horizontal coordinate of the lower left pixel point is taken as the third difference, and the ratio of the third difference to the second difference is multiplied by the pixel value of the lower right pixel point to obtain a second product, and the first product is added to the second product to obtain a first pixel filling value of the edge pixel point; The difference between the abscissa of the upper right pixel point and the abscissa of the edge pixel point and the abscissa of the upper left pixel point is taken as the fourth difference and the fifth difference, and the ratio of the fourth difference to the fifth difference is multiplied by the pixel value of the upper left pixel point to obtain a third product; the difference between the abscissa of the edge pixel point and the abscissa of the upper left pixel point is taken as the sixth difference, and the ratio of the sixth difference to the fifth difference is multiplied by the pixel value of the upper right pixel point to obtain a fourth product, and the third product is added to the fourth product to obtain the second pixel filling value of the edge pixel point; The difference between the ordinate of the upper right pixel point and the ordinate of the edge pixel point and the ordinate of the lower right pixel point is taken as the seventh difference and the eighth difference, and the ratio of the seventh difference to the eighth difference is multiplied by the first pixel filling value to obtain a fifth product; the difference between the ordinate of the edge pixel point and the ordinate of the lower left pixel point is taken as the ninth difference, and the ratio of the ninth difference to the eighth difference is multiplied by the second pixel filling value to obtain a sixth product, and the fifth product is added to the sixth product to obtain the pixel filling value of the edge pixel point; The above operations are performed on all pixels in the blank area in order from the edge to the inside to complete the filling of the blank area, and the second Radon filling area corresponding to the second Radon correction area is obtained; Performing an inverse Radon transform on the second Radon filled area to obtain a de-artifacted background area after the artifacts are removed; The image region composed of the pathological region and the artifact-removed background region is used as a pre-processed pathological slice image corresponding to the pathological slice image to be processed after artifacts are removed; The above operation is performed on a plurality of enhanced pathological slice images to obtain a plurality of pre-processed pathological slice images corresponding to the plurality of enhanced pathological slice images after the artifact removal operation is performed on the plurality of enhanced pathological slice images.

6. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 2, characterized in that: The plurality of pre-processed pathological slice images are divided to obtain a plurality of slice image groups, including: Obtaining the number of pre-processed pathological slice images and the preset slice image group division parameters; Determine the number of slice image groups after the division according to the number of images and the division parameters; According to the number M of preprocessed pathology slice images with benign diagnosis results, the number N of preprocessed pathology slice images with malignant diagnosis results and the number of slice image groups, the number M of preprocessed pathology slice images with benign diagnosis results and the number N of preprocessed pathology slice images with malignant diagnosis results contained in each slice image group are determined; wherein M:N=1:1; According to the number M of benign pre-processed pathology slice images and the number N of malignant pre-processed pathology slice images contained in each slice image group, a corresponding pre-processed pathology slice image is allocated to each slice image group, and a number of slice image groups are determined according to the allocation results.

7. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 1, characterized in that: According to the pathological identification result corresponding to each preprocessed pathological slice image in each slice image group and the collected corresponding pathological diagnosis result, a first evaluation index and a second evaluation index corresponding to each slice image group are generated, including: Select any slice image group as the slice image group to be processed; The number of pre-processed pathological slice images with benign pathological identification results in the group of slice images to be processed is recorded as G, and the number of pre-processed pathological slice images with malignant pathological identification results is recorded as B; Compare the pathological recognition results in the to-be-processed slice image group with the corresponding collected pathological diagnosis results, record the number of pre-processed pathological slice images whose pathological recognition results and the corresponding collected pathological diagnosis results are both benign as G1, and record the number of pre-processed pathological slice images whose pathological recognition results and the corresponding collected pathological diagnosis results are both malignant as B1; The number of pre-processed pathological sections whose pathological identification results are malignant but whose corresponding pathological diagnosis results are benign is recorded as G2; According to the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all benign, the number of pre-processed pathology slice images whose corresponding pathological diagnosis results are all malignant, the number of pre-processed pathology slice images whose pathological identification results are benign in the slice image group to be processed, and the number of pre-processed pathology slice images whose pathological identification results are malignant, the diagnostic accuracy is calculated as the first evaluation indicator corresponding to the slice image group to be processed; Based on the number of preprocessed pathology slice images whose pathology recognition results and the corresponding collected pathology diagnosis results are both benign and the number of preprocessed pathology slice images whose pathology recognition results are malignant but the corresponding collected pathology diagnosis results are benign, the diagnostic sensitivity is calculated as the second evaluation indicator corresponding to the group of slice images to be processed.

8. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 1, characterized in that: The third evaluation index corresponding to each slice image group is generated according to the model diagnosis time corresponding to each pre-processed pathological slice image contained in each slice image group, including: Select any slice image group as the slice image group to be processed; Determine the average diagnosis time of the group of slice images to be processed according to the model diagnosis time corresponding to each pre-processed pathological slice image contained in the group of slice images to be processed; The average diagnosis time is used as the third evaluation indicator corresponding to the slice image group to be processed.

9. The evaluation method of the pathology recognition model applied to pathology diagnosis according to claim 1, characterized in that: The performance of the pathology recognition model to be evaluated is evaluated according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group, including: Determine the relationship between the first evaluation indicator and the first preset threshold, the second evaluation indicator and the second preset threshold, and the third evaluation indicator and the third preset threshold; When all slice image groups meet the conditions that the first evaluation index is higher than the first preset threshold, the second evaluation index is higher than the second preset threshold, and the third evaluation index is lower than the third preset threshold, it is determined that the performance evaluation of the pathology recognition model to be evaluated is qualified; otherwise, it is determined that the performance evaluation of the pathology recognition model to be evaluated is unqualified.

10. An evaluation device for a pathology recognition model used in pathology diagnosis, characterized in that: include: An information collection module is used to collect a number of pathological slice images and a number of corresponding pathological diagnosis results; An image preprocessing module, used for performing image preprocessing on the plurality of pathological slice images to obtain a plurality of preprocessed pathological slice images; An image division module, used for dividing the plurality of pre-processed pathological slice images to obtain a plurality of slice image groups; A pathological diagnosis module, used to sequentially input a plurality of pre-processed pathological slice images contained in each slice image group into the pathological recognition model to be evaluated, and obtain a pathological recognition result and a model diagnosis time corresponding to each pre-processed pathological slice image in each slice image group; Indicator generation module for: Generate a first evaluation index and a second evaluation index corresponding to each slice image group according to the pathology recognition result corresponding to each preprocessed pathology slice image in each slice image group and the collected corresponding pathology diagnosis result; Generate a third evaluation index corresponding to each slice image group according to the model diagnosis time corresponding to each preprocessed pathological slice image included in each slice image group; The performance evaluation module is used to evaluate the performance of the pathology recognition model according to the first evaluation index, the second evaluation index and the third evaluation index corresponding to each slice image group.

Images