Histopathologic identification method and device for prostate cancer lymph node metastasis

Abstract

The invention discloses a histopathologic recognition method and device for prostate cancer lymph node metastasis, and the method comprises the steps: carrying out the pathology recognition according to a pathology image of a prostate lymph node, and obtaining a first confidence coefficient corresponding to an original pathology recognition result; performing mask processing on the pathological image to obtain an original pathological mask image; correcting the first confidence coefficient according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence coefficient; and when the original pathological mask image comprises the target tissue mask part, performing pixel correction on the target tissue mask part according to the target confidence, and outputting a target pathological mask image. The first confidence coefficient is corrected through the plurality of pathological example images in the preset pathological image library, so that confidence coefficient errors caused by insufficient generalization performance of pathological recognition can be reduced, and the accuracy of the output target pathological mask image can be improved.

Description

Technical Field

[0001] The present application relates to the field of medical technology, and in particular to a method and device for histopathological identification of prostate cancer lymph node metastasis. Background Art

[0002] Prostate cancer often metastasizes to lymph nodes. At present, when treating prostate cancer patients, it is often necessary to perform a pelvic lymph node dissection at the same time as radical prostatectomy, and perform a histopathological section examination on the removed lymph nodes to observe whether there are cancer cells in the lymph node tissue to determine whether prostate cancer has metastasized to the lymph nodes, thereby determining the next treatment strategy. However, the pathological reading of lymph nodes is a monotonous and arduous visual recognition task, and the human eye is easily fatigued and easily misses lymph node metastases with small diameters. The relative shortage of pathologists nationwide and the heavy workload of pathologists have exacerbated this phenomenon. Previous studies have shown that the misdiagnosis rate of prostate cancer lymph node micrometastases (<2mm) in conventional pathology reports is as high as 8~13%. This greatly affects the quality of diagnosis and treatment of prostate cancer patients.

[0003] In order to improve the accuracy of pathological diagnosis of prostate cancer lymph node metastasis and reduce the missed diagnosis rate of prostate cancer lymph node micrometastasis lesions, image recognition technology can be introduced to identify prostate lymph node pathological tissue to help pathologists perform pathological observation. However, in the prior art, there are few recognition technologies for lymph node pathological tissue, and the existing recognition technologies for lymph node pathological tissue are relatively simple and have average accuracy. When using the existing lymph node pathological tissue recognition technology to perform lymph node pathological film reading and recognition of prostate cancer lymph node metastasis lesions, it is easy to miss the diagnosis. Summary of the invention

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The embodiments of the present application provide a method and device for histopathological identification of prostate cancer lymph node metastasis, which can improve the accuracy of identifying prostate lymph node pathological tissue.

[0006] On the one hand, the present application embodiment provides a method for histopathological identification of prostate cancer lymph node metastasis, comprising the following steps: Performing pathological recognition according to the pathological image of the prostate lymph node to obtain a first confidence level corresponding to the original pathological recognition result; Performing mask processing on the pathological image to obtain an original pathological mask image; Correcting the first confidence level according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence level; When the original pathology mask image includes a target tissue mask portion, pixel correction is performed on the target tissue mask portion according to the target confidence level, and a target pathology mask image is output.

[0007] Optionally, the plurality of pathological example images form a plurality of pathological image classification clusters in the preset pathological image library; The performing pathology recognition according to the pathology image to obtain a first confidence level corresponding to the original pathology recognition result includes: Extracting features from the pathological image to obtain a transitional pathological feature image; The step of correcting the first confidence level according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence level includes: Clustering the transition pathology feature images according to the plurality of pathology image classification clusters, and determining a target pathology image classification cluster from the plurality of pathology image classification clusters, wherein the target pathology image classification cluster includes the plurality of pathology example images; The first confidence is corrected according to a plurality of the pathological example images in the target pathological image classification cluster to obtain the target confidence.

[0008] Optionally, the preset pathological image library includes a pathological image classification label corresponding to each pathological image classification cluster, and a second confidence level corresponding to each pathological example image, wherein the second confidence level is a pathological tissue recognition confidence level obtained by performing pathological recognition based on the pathological example image; The step of correcting the first confidence level according to the plurality of pathological example images in the target pathological image classification cluster to obtain the target confidence level includes: According to the pathological image classification label of the target pathological image classification cluster, determining a target preset correction offset value corresponding to the target pathological image classification cluster from a plurality of preset correction offset values, wherein the plurality of preset correction offset values ​​correspond one-to-one to the plurality of pathological image classification clusters; The first confidence is corrected according to the plurality of second confidences corresponding to the target pathological image classification cluster and the target correction offset value to obtain the target confidence.

[0009] Optionally, the first confidence is corrected according to the plurality of second confidences corresponding to the target pathological image classification cluster and the target correction offset value to obtain the target confidence, comprising: Determining a third confidence level and a fourth confidence level among the plurality of second confidence levels corresponding to the target pathological image classification cluster; The first confidence is corrected according to the third confidence, the fourth confidence and the target correction offset value to obtain the target confidence.

[0010] Optionally, performing pixel correction on the target tissue mask portion according to the target confidence and outputting a target pathology mask image comprises: determining a target pathology recognition result according to the target confidence; When the target pathology recognition result indicates that the prostate lymph node includes the target tissue, a plurality of target pixels are generated according to the target confidence and the target tissue mask portion; The target tissue mask portion is pixel-replaced according to the plurality of target pixels, and the target pathology mask image is output.

[0011] Optionally, the target tissue mask portion includes a plurality of original pixels; The generating a plurality of target pixels according to the target confidence and the target tissue mask portion comprises: For each of the original pixels, adjusting the pixel value of the original pixel according to the target confidence level to obtain a target pixel value corresponding to the original pixel; A plurality of target pixels are correspondingly generated according to the plurality of target pixel values.

[0012] Optionally, the pathological image is obtained by acquiring a pathological slice image of a prostate lymph node and performing a sliding window cropping process on the pathological slice image, wherein the pathological image obtained by performing a sliding window cropping process on the pathological slice image is a plurality of pathological images; Said also includes: Performing pathology recognition on the next pathology image to output the target pathology mask image corresponding to the next pathology image; After obtaining the target pathology mask images corresponding to the multiple pathology images, graphics are merged according to the target pathology mask images corresponding to the multiple pathology images to obtain the target pathology slice mask image corresponding to the pathology slice image, wherein the target pathology slice mask image includes a target tissue body mask of the prostate lymph node.

[0013] On the other hand, the present application also provides a histopathological identification device for prostate cancer lymph node metastasis, comprising: The original pathology recognition module is used to perform pathology recognition based on the pathology image of the prostate lymph node and obtain a first confidence level corresponding to the original pathology recognition result; An original pathology mask processing module, used for performing mask processing on the pathology image to obtain an original pathology mask image; A target confidence acquisition module, used for correcting the first confidence according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence; The target pathology mask image output module is used to perform pixel correction on the target tissue mask part according to the target confidence when the original pathology mask image includes the target tissue mask part, and output the target pathology mask image.

[0014] On the other hand, an embodiment of the present application further provides an electronic device, including: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the aforementioned method is implemented.

[0015] On the other hand, an embodiment of the present application further provides a computer-readable storage medium, which stores a computer program executable by a processor, and the processor-executable computer program is used to implement the method described above when executed by the processor.

[0016] On the other hand, an embodiment of the present application also provides a computer program product, including a computer program or computer instructions, wherein the computer program or the computer instructions are stored in a computer-readable storage medium, and a processor of an electronic device reads the computer program or the computer instructions from the computer-readable storage medium, and the processor executes the computer program or the computer instructions, so that the electronic device performs the method described above.

[0017] The embodiments of the present application include at least the following beneficial effects: pathology recognition can be first performed based on the pathology image of the prostate lymph nodes to obtain a first confidence level corresponding to the original pathology recognition result; then, the pathology image is masked to obtain an original pathology mask image; then, the first confidence level is corrected based on a plurality of pathology example images in a preset pathology image library to obtain a target confidence level; when the original pathology mask image includes a target tissue mask portion, pixel correction is performed on the target tissue mask portion based on the target confidence level to output a target pathology mask image. By correcting the first confidence by using multiple pathological example images in a preset pathological image library, the error in the confidence corresponding to the difficult-to-identify lymph node metastasis caused by insufficient generalization performance of pathological recognition can be reduced, so that the target confidence can be closer to the actual recognition result, and the difficult-to-identify lymph node metastasis can be more accurately identified, and it can be used as a reliable reference for the original pathological mask image. Then, the target tissue mask part is pixel-corrected by the target confidence so that the target pathological mask image can correspond to the target confidence, so that the target pathological mask image can more accurately display the difficult-to-identify lymph node metastasis, and then the target pathological mask image can be closer to the actual recognition result, providing more accurate recognition results for medical staff, and reducing the need for medical staff to spend extra time to determine whether the recognition results of prostate lymph node pathological tissue are correct.

[0018] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained through the structures particularly pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide further understanding of the technical solution of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present application and do not constitute a limitation on the technical solution of the present application.

[0020] Figure 1 It is a flow chart of a method for histopathological identification of prostate lymph node metastasis provided in an embodiment of the present application; Figure 2 It is an execution framework of a method for histopathological identification of prostate lymph node metastasis provided in a specific embodiment of the present application; Figure 3 It is an execution framework of a method for histopathological identification of prostate lymph node metastasis provided by another specific embodiment of the present application; Figure 4 is a schematic diagram of the enhanced target tissue mask portion provided in an embodiment of the present application; Figure 5 is a schematic diagram of deleting a target tissue mask portion provided by an embodiment of the present application; Figure 6 yes Figure 1 A flowchart of sub-steps of a specific embodiment of step 103; Figure 7 yes Figure 6 A flowchart of sub-steps of a specific embodiment of step 602; Figure 8 yes Figure 7 A flowchart of sub-steps of a specific embodiment of step 702; Fig. 9 yes Figure 1 A flowchart of sub-steps of a specific embodiment of step 104; Fig.10 yes Fig. 9 A flowchart of sub-steps of a specific embodiment of step 902; Fig.11 is a schematic diagram of a tissue pathology identification device provided in an embodiment of the present application; Fig.12 It is a schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0022] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0024] In order to improve the accuracy of pathological diagnosis of prostate cancer lymph node metastasis and reduce the missed diagnosis rate of prostate cancer lymph node micrometastasis lesions, image recognition technology can be introduced to identify prostate lymph node pathological tissue to help pathologists perform pathological observation. However, in the prior art, there are few recognition technologies for lymph node pathological tissue, and the existing recognition technologies for lymph node pathological tissue are relatively simple and have average accuracy. When using the existing lymph node pathological tissue recognition technology to perform lymph node pathological film reading and recognition of prostate cancer lymph node metastasis lesions, it is easy to miss the diagnosis.

[0025] In order to improve the recognition accuracy of prostate lymph node pathological tissue, the embodiments of the present application provide a histopathological recognition method for prostate lymph node metastasis, a histopathological recognition device for prostate lymph node metastasis, an electronic device, a computer-readable storage medium and a computer program product. The histopathological recognition method can first perform pathological recognition based on a pathological image of the prostate lymph node to obtain a first confidence level corresponding to the original pathological recognition result, then perform mask processing on the pathological image to obtain an original pathological mask image, and then correct the first confidence level based on multiple pathological example images in a preset pathological image library to obtain a target confidence level. When the original pathological mask image includes a target tissue mask part, pixel correction is performed on the target tissue mask part based on the target confidence level to output the target pathological mask image. By correcting the first confidence by using multiple pathological example images in a preset pathological image library, the error in the confidence corresponding to the difficult-to-identify lymph node metastasis caused by insufficient generalization performance of pathological recognition can be reduced, so that the target confidence can be closer to the actual recognition result, and the difficult-to-identify lymph node metastasis can be more accurately identified, and it can be used as a reliable reference for the original pathological mask image. Then, the target tissue mask part is pixel-corrected by the target confidence so that the target pathological mask image can correspond to the target confidence, so that the target pathological mask image can more accurately display the difficult-to-identify lymph node metastasis, and then the target pathological mask image can be closer to the actual recognition result, providing more accurate recognition results for medical staff, and reducing the need for medical staff to spend extra time to determine whether the recognition results of prostate lymph node pathological tissue are correct.

[0026] Reference Figure 1 As shown, Figure 1 This is a flowchart of a method for histopathological identification of prostate lymph node metastasis provided in an embodiment of the present application. The method for histopathological identification can be executed by a server, or by an electronic observation device such as an electron microscope, or by an electronic observation device and a server. In the embodiment of the present application, the method is executed by a server as an example for explanation, and other execution examples can refer to the following embodiments. In the embodiment of the present application, the method for histopathological identification of prostate lymph node metastasis includes but is not limited to the following steps.

[0027] Step 101: performing pathological recognition according to the pathological image of the prostate lymph node to obtain a first confidence level corresponding to the original pathological recognition result; Step 102: performing mask processing on the pathological image to obtain an original pathological mask image; Step 103: correcting the first confidence level according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence level; Step 104: When the original pathology mask image includes the target tissue mask part, pixel correction is performed on the target tissue mask part according to the target confidence level, and a target pathology mask image is output.

[0028] In one embodiment, the pathological image of the prostate lymph node is obtained by acquisition, wherein the acquisition method may be imaging the pathological section through an electrical observation device, receiving the pathological image sent by the electrical observation device, or calling the pathological image of the prostate lymph node to be processed that is uploaded and stored in a database, etc., which is not specifically limited here. In addition, the pathological image of the prostate lymph node may be a pathological section image of the prostate lymph node, or a part of the pathological section image, which is not specifically limited here.

[0029] In addition, when the pathological image of the prostate lymph node is part of the pathological slice image, the pathological image can be obtained in the overall pathological slice image by the following method, or in the pathological slice image by other methods, which are not specifically limited here. For example, one or more regions of interest are identified on the pathological slice image by region of interest (ROI) recognition, and the image is cut according to the identified region of interest as the target, and the image obtained by cutting is the pathological image. For another example, the target area is determined by obtaining external frame selection information. Specifically, when the medical staff is interested in a part of the pathological slice image, the region of interest can be framed by frame selection. The device will form external frame selection information according to the framed area, and the pathological slice image is cut according to the external frame selection, and the image obtained is the pathological image. For another example, all pathological slice images are cut according to the preset cutting size, and the images obtained by cutting are all pathological images.

[0030] In addition, the obtained pathological image can be an original image without image preprocessing, or an original pathological image after image preprocessing, which is not specifically limited here. Among them, the image preprocessing can be image filtering, image smoothing, grayscale, etc., which is not specifically limited here.

[0031] In addition, the original pathological recognition result refers to the recognition result finally output by the pathological recognition in step 101, which can represent one or more target tissues on the pathological image. For example, if there is only one target tissue in the prostate lymph node on the pathological image, the original pathological recognition result can represent the target tissue; for another example, if there are multiple target tissues in the prostate lymph node on the pathological image, the original pathological recognition result can represent multiple target tissues. In the process of pathological recognition, multiple classifications will be performed to obtain multiple classification confidences. The recognition result finally output is the result corresponding to the maximum confidence among the multiple classification confidences (that is, the original pathological recognition result), and the confidence corresponding to the recognition result finally output is the first confidence corresponding to the original pathological recognition result. Among them, the original pathological recognition result can be a two-classification result, a three-classification result, a four-classification result, etc., which is not specifically limited here. In addition, the number of first confidences corresponds to the number of target tissues represented by the original pathological recognition result, that is, each identified target tissue corresponds to a first confidence. For the convenience of the subsequent example description, the original pathology recognition result is a binary classification result, which is used to indicate whether the prostate lymph node includes the target tissue, the prostate lymph node on the pathology image has only one target tissue, and the first confidence level has only one example for the following example description. Other categories, such as prostate lymph nodes have multiple target tissues, multiple first confidence levels, etc., can be executed one by one with reference to the following examples, and no further elaboration will be given here.

[0032] In one embodiment, the target tissue may be a pathological tissue on the prostate lymph node, or may be a healthy tissue on the prostate lymph node, which is not specifically limited here. In order to facilitate the description of the subsequent embodiments, the subsequent embodiments are further described by taking the target tissue as a pathological tissue as an example (that is, the original pathological recognition result is used to indicate whether the prostate lymph node has a pathological tissue). The embodiment in which the target tissue is a healthy tissue can be replaced with reference to the following embodiment, and will not be described in detail later.

[0033] In one embodiment, pathological identification can be achieved in the following manner: first input the pathological image of the prostate lymph nodes into the first network. The first network can first perform feature extraction on the pathological image of the prostate lymph nodes to obtain a first feature image of the pathological image, and then perform feature processing such as pooling and deep convolution based on the first feature image to normalize the information in the first feature image to obtain low-dimensional first feature information, and calculate the classification confidence of the first feature information through an activation function to obtain multiple classification confidences, and use the recognition result corresponding to the classification confidence with the highest value among the multiple classification confidences as the original pathological identification result, and use the highest classification confidence as the first confidence. Among them, the first network can be a convolutional neural network (CNN), a convolutional recurrent neural network (CRNN), etc., which is not specifically limited here.

[0034] In one embodiment, the mask processing can be implemented in the following manner: first input the pathological image of the prostate lymph node into the second network. The second network can first perform feature extraction on the pathological image of the prostate lymph node to obtain a second feature image of the pathological image of the prostate lymph node, and then upsample the second feature image to perform pixel prediction, and then classify the pixels in the feature image after upsampling through multiple convolutional layers, and generate mask information on the pathological image according to the pixel classification results, so as to obtain the original pathological mask image. Among them, the second network can be a U-net network (U-Net), a deep label network (DeepLab), etc., which are not specifically limited here. In addition, when the mask processing can classify the pixels related to the target tissue, the original pathological mask image can carry the target tissue mask information and other mask information such as background mask information; when the mask processing cannot classify the pixels related to the target tissue, the original pathological mask image can carry other mask information such as background mask information, which is not specifically limited here.

[0035] In one embodiment, the preset pathological image library may be a common database or a database combining common data storage function and vector storage function, which is not specifically limited here. In addition, the pathological example image may be a typical pathological image of a prostate lymph node or a feature image of a typical pathological image of a prostate lymph node after feature extraction, which is not specifically limited here. Among them, the typical pathological images mentioned may be pathological images of target tissues with a relatively large area, pathological images of relatively concentrated target tissues, pathological images of target tissues with better staining effects, pathological images with obvious morphological differences between cells in the target tissue and normal cells, pathological images with obvious morphological differences between the nuclei of cells in the target tissue and the nuclei of normal cells, pathological images with disordered cell arrangement in the target tissue, etc., which belong to easy sample pathological images; may also be pathological images of target tissues with a relatively small area, pathological images of relatively dispersed target tissues, pathological images of target tissues with poor staining effects, pathological images with small morphological differences between cells in the target tissue and normal cells under specific circumstances, pathological images with small morphological differences between the nuclei of cells in the target tissue and the nuclei of normal cells under specific circumstances, pathological images with small arrangement structure differences between the cell arrangement structure of the target tissue and the cell arrangement structure of normal tissue under specific circumstances, etc., which belong to hard sample pathological images, and are not specifically limited here.

[0036] Reference Figure 2 As shown, Figure 2 The execution architecture of the histopathological identification method of prostate lymph node metastasis in a specific embodiment is shown. In one embodiment, step 101 and step 102 use parallel networks, and pathological identification and mask processing use feature extraction methods of different precisions respectively. Referring to the implementation methods of the above-mentioned pathological identification and mask processing, step 101 uses the first feature image to complete the pathological identification process and obtain the first confidence, and step 102 uses the second feature image to complete the mask processing process and obtain the original pathological mask graphic. Different image processing tasks require different image information accuracies. If the same feature image is used, for a certain execution network, image information may be missing, thereby affecting the execution accuracy of at least one of pathological identification and mask processing. By using an execution architecture with a parallel network, pathological identification and mask processing can obtain image information (i.e., feature images) of the required accuracy, so that the execution networks of pathological identification and mask processing can output the first confidence with higher reliability and the original pathological mask image with higher accuracy. Then, a more accurate target pathological mask image can be output later, thereby improving the efficiency of medical staff in pathological judgment of prostate lymph nodes.

[0037] Reference Figure 3 As shown, Figure 3 FIG. 2 shows an execution architecture of a method for histopathological identification of prostate lymph node metastasis in another specific embodiment. In one embodiment, the network parts used in step 101 and step 102 are merged. Specifically, Figure 3 In the execution architecture shown, the pathological image can be firstly feature extracted to obtain a transitional pathological feature image. For mask processing, upsampling can be performed based on the transitional pathological feature image, and pixel prediction can be performed by upsampling to increase the receptive field, and then each pixel in the upsampled transitional pathological feature image is classified through multiple convolutional layers, and a mask is formed on the pathological image according to the classification of each pixel to obtain the original pathological mask image; and for pathological recognition, the number of features of the transitional pathological feature image can be unified to obtain a third feature image, and then the healthy tissue and pathological tissue of the prostate lymph node on the pathological image can be classified according to the third feature image to obtain the classification confidence of each classification, and the maximum classification confidence is used to determine the first confidence corresponding to the original pathological recognition result. Generally speaking, compared with pathology recognition, mask processing has lower requirements for the accuracy of image information. Therefore, based on the accuracy requirements of the feature image of mask processing, a transition feature image can be first extracted. At this time, the transition feature image meets the accuracy of the steps of upsampling and outputting the original pathology mask image in mask processing. Since the accuracy requirements of the feature image of pathology recognition are relatively high, feature extraction can be performed on the basis of the transition pathology feature image to obtain a third pathology feature image with higher accuracy, so as to meet the image information accuracy requirements required for pathology recognition. By adopting Figure 3 The combined execution structure shown in enables the device to perform multi-level feature extraction according to the accuracy requirements of different image processing tasks to be performed, thereby reducing the computing resources used in the device's feature information extraction steps, so that the efficiency of the pathological tissue identification process can be improved.

[0038] In order to facilitate the description of the following content, the following embodiments will be based on Figure 3 The execution architecture in Figure 2 The implementation example of the execution structure in can refer to the following implementation example and will not be described in detail here.

[0039] In one embodiment, the pathology example image is a historical transition pathology feature image after step 101 to step 104. Each pathology example image has a corresponding first evaluation score and a second evaluation score stored in a preset pathology image library. Among them, the first evaluation score can be used to evaluate the gap between the target pathology mask image corresponding to the pathology example image and the original pathology mask image, and the second evaluation score can be used to evaluate the gap between the target pathology mask image of the pathology example image and the real pathology mask image. The real pathology mask image is an artificial pathology mask image corrected by medical staff on the basis of the target pathology mask image. If the medical staff does not correct the target pathology mask image, the second evaluation score is 0. Specifically, after each pathological image has gone through steps 101 to 104, a first evaluation score and a second evaluation score corresponding to the pathological image are calculated based on the transition pathological feature image, and then the transition pathological feature image is stored as a pathological example image in a preset pathological image library, and its first evaluation score and second evaluation score are stored correspondingly in the preset pathological image library; in the process of generating a target pathological mask image corresponding to a new pathological image, a plurality of pathological example images are randomly selected from the preset pathological image library, and the first evaluation scores and second evaluation scores corresponding to these pathological example images are determined, and the first confidence of the new pathological image is corrected based on these first evaluation scores and second evaluation scores to obtain the target confidence.

[0040] In one embodiment, in step 103, one or more similar pathology example images are determined from multiple pathology example images based on the transition feature image, and the first confidence level is lowered or increased based on whether these pathology example images indicate that prostate lymph node tissue includes the target tissue, thereby obtaining a target confidence level.

[0041] In the pathological sections of prostate lymph nodes, the target tissue may have different morphologies, dispersed positions, poor staining effects, and there is a large difference between the color distribution domain corresponding to the data set used by the execution architecture during training and the color distribution domain of the pathological image during actual pathological reading. Therefore, the original pathological recognition result may not be accurate (that is, the first confidence is unreliable), resulting in the original pathological mask image not corresponding to the original pathological recognition result, which makes the accuracy of the original pathological mask image questionable. For example, the target tissue occupies a small area in the prostate lymph node tissue, and the target tissue and the surrounding healthy tissue are not highly distinguished in the pathological image. This information may be ignored in the pathological recognition process due to deep convolution and other reasons, and ultimately an erroneous original pathological recognition result may be obtained. Since the transition feature processing image is upsampled in the mask processing, the information of distinguishing the target tissue from the surrounding healthy tissue can be enhanced, so that the original mask image can accurately show the target tissue included in the pathological image through the target tissue mask part carried. These situations will make the original mask image contradict the original pathological recognition result obtained previously, but reduce the credibility of the original pathological mask image. For another example, due to uneven staining and different imaging parameters used by medical staff when performing imaging operations on slices, the imaging effects of tissues on pathological images are different, resulting in a large difference between the color distribution domain of the pathological image and the color distribution domain corresponding to the data set used in the training of the execution architecture, ultimately leading to erroneous original pathological results and erroneous original pathological mask images, reducing the reliability of pathological recognition.

[0042] To this end, the first confidence level can be corrected by using similar pathological example images to improve the accuracy of the original pathological recognition result and obtain a more accurate pathological recognition result (i.e., the target pathological recognition result that appears below). Relatively speaking, the reliability of the corrected target confidence level is also higher than the first confidence level. Therefore, by correcting the first confidence level by using multiple pathological example images in the preset pathological image library, the accuracy of the original pathological recognition result can be corrected, so that the recognition error caused by the error between the training set and the actual image in the execution architecture can be reduced, thereby reducing the impact caused by the insufficient generalization performance of pathological recognition, and providing a reference for the accuracy of the original pathological mask image, so that the original mask pathological image can be adjusted accordingly through the target confidence level, so that the target tissue that is difficult to identify on the pathological image can be correctly displayed on the target pathological mask image, so that the final output target pathological mask image can be more accurate.

[0043] In one embodiment, referring to Figure 4As shown, in step 104, when the original pathology mask image includes a target tissue mask portion, and the target confidence indicates that the prostate lymph node includes the target tissue, the target tissue mask portion is pixel-corrected to enhance the target tissue mask information in the original pathology mask image, so that the first confidence corresponding to the target tissue in the output target pathology mask image can be improved, thereby allowing medical staff to more easily and clearly see the target tissue in the prostate lymph node in the pathology image.

[0044] In one embodiment, referring to Figure 5 As shown, in step 104, when the original pathology mask image includes the target tissue mask portion, and the target confidence indicates that the target tissue is not included in the prostate lymph node, the pixel value corresponding to the target tissue mask information in the original pathology mask image is reduced by performing pixel correction on the target tissue mask portion, so as to correct the prediction result of the portion of the original pathology mask image that was previously predicted as the target tissue, so that these portions are not predicted as the target tissue, thereby making it possible to make the output target pathology mask image not carry erroneous target tissue mask information, thereby avoiding the need for medical staff to spend extra time on manual judgment of the target tissue mask portion to a certain extent.

[0045] In one embodiment, when the original pathology mask image does not include the target tissue mask portion, and the target confidence indicates that the prostate lymph nodes do not include the target tissue, the original pathology mask image is output as the target pathology mask image.

[0046] In one embodiment, when the original pathological mask image does not include the target tissue mask portion, and the target confidence indicates that the prostate lymph node includes the target tissue, the current pathological image can be cropped into multiple pathological sub-images according to the preset image size by using image cropping methods such as sliding window cropping. Mask processing is performed on each pathological sub-image to obtain the original pathological mask sub-image corresponding to each pathological sub-image, pixel correction is performed on each original pathological mask sub-image according to the target confidence, and the target pathological mask sub-image corresponding to each pathological sub-image is output, and multiple target pathological mask sub-images are merged to obtain the target pathological mask image. The merging may include image synthesis processing, image filtering processing, image edge smoothing processing, etc., which are not specifically limited here.

[0047] In one embodiment, the preset pathological image library may store the correct recognition results corresponding to each pathological example image, and step 103 may be implemented in the following manner: respectively calculating the similarity between the pathological image and each pathological example image, ranking each pathological example image in descending order according to the similarity, and determining the top K pathological example images as target pathological example images, where K is a natural odd number greater than 0. Determine the proportion of recognition results corresponding to the K target pathological example images, determine the specific recognition result corresponding to one with the larger recognition result proportion as a correction label, and correct the first confidence by the highest similarity corresponding to the correction label and the correction label itself to obtain the target confidence.

[0048] Reference Figure 6 As shown, Figure 6 Shows Figure 1 In one embodiment, a plurality of pathological example images form a plurality of pathological image classification clusters in a preset pathological image library, and step 103 may include but is not limited to the following steps.

[0049] Step 601: clustering the transition pathology feature image according to a plurality of pathology image classification clusters, and determining a target pathology image classification cluster from the plurality of pathology image classification clusters; Step 602: Correct the first confidence level according to a plurality of pathological example images in the target pathological image classification cluster to obtain a target confidence level.

[0050] In one embodiment, multiple pathological example images can form multiple pathological image classification clusters in a preset pathological image library in the following manner, or can form multiple pathological image classification clusters in other manners, which are not specifically limited here. For example, the similarity between each pathological example image and other pathological example images is calculated, and each pathological example image is classified according to the similarity between each pathological example image and other pathological example images to obtain multiple pathological image classification clusters. For another example, multiple pathological image classification clusters are obtained by classifying the attributes of each pathological example image, among which the attributes can be whether the prostate lymph nodes in the pathological example image have target tissues, or whether the prostate lymph nodes in the pathological example image have specific types of target tissues, etc., which are not specifically limited here. In addition, the target pathological image classification cluster includes multiple pathological example images.

[0051] In one embodiment, a pathological image classification cluster is formed based on the similarity between multiple pathological example images, and clustering of transitional pathological feature images can be achieved in the following manner: first, the similarity between the transitional pathological feature image and each pathological example image is calculated, and the pathological example image most similar to the transitional pathological feature image is determined, and the pathological image classification cluster corresponding to the most similar pathological example image is determined as the target pathological image classification cluster corresponding to the transitional pathological feature image.

[0052] In one embodiment, a pathological image classification cluster is formed by clustering multiple pathological example images based on a certain clustering algorithm. Under this clustering method, the prostate lymph nodes in each pathological example image in each pathological image classification cluster may include the target tissue or may not include the target tissue, etc., which is not specifically limited here. Among them, the clustering algorithm can be a K-means clustering algorithm (K-means), a hierarchical clustering algorithm (Hierarchical Clustering), etc., which is not specifically limited here. Clustering of transitional pathological feature images can be achieved in the following manner: using the same clustering algorithm to divide the transitional pathological feature images to classify them into one of the multiple pathological image classification clusters, and the pathological image classification cluster is the target pathological image classification cluster.

[0053] In one embodiment, a preset pathological image library may store a preset compensation value and a corresponding correct recognition result corresponding to each pathological example image, and a pathological image classification cluster is formed based on the similarity between multiple pathological example images. Step 602 may determine the target confidence in the following manner: first determine the pathological example image with the highest similarity among multiple pathological example images in the target pathological image classification cluster, then determine whether the correct recognition result corresponding to the pathological example image corresponds to the original pathological recognition result. If the correct recognition result corresponding to the pathological example image indicates that the corresponding prostate lymph node does not include the target tissue, while the original pathological recognition result indicates that the corresponding prostate lymph node includes the target tissue, a penalty correction is performed on the first confidence according to the preset compensation value to obtain the target confidence. If the correct recognition result corresponding to the pathological example image indicates that the corresponding prostate lymph node includes the target tissue, while the original pathological recognition result indicates that the corresponding prostate lymph node includes the target tissue, then the target confidence is obtained. If the corresponding prostate lymph node also includes the target tissue, the first confidence level is rewarded and corrected according to the preset compensation value to obtain the target confidence level; if the correct recognition result corresponding to the pathological example image indicates that the corresponding prostate lymph node includes the target tissue, and the original pathological recognition result indicates that the corresponding prostate lymph node does not include the target tissue, the first confidence level is penalty-corrected according to the preset compensation value to obtain the target confidence level; if the correct recognition result corresponding to the pathological example image indicates that the corresponding prostate lymph node does not include the target tissue, and the original pathological recognition result indicates that the corresponding prostate lymph node does not include the target tissue, the first confidence level is rewarded and corrected according to the preset compensation value to obtain the target confidence level.

[0054] In one embodiment, a preset pathology image library may store the similarity between each pathology example image and other pathology example images, and a pathology image classification cluster is formed based on the attribute of whether multiple pathology example images include a target tissue portion. Step 602 may determine the target confidence in the following manner: first determine the pathology example image with the highest similarity to the transition pathology feature image and its specific similarity among multiple pathology example images in the target pathology image classification cluster, and determine whether the attribute of the target pathology image classification cluster corresponds to the original pathology recognition result. If the attribute corresponding to the target pathology image classification cluster is that the pathology example image includes the target tissue portion, and the original pathology recognition result indicates that the corresponding prostate lymph node includes the target tissue, the first confidence is increased according to the similarity between the transition pathology feature image and the most similar pathology example image to obtain the target confidence. If the attribute corresponding to the target pathology image classification cluster is that the pathology example image does not include the target tissue portion, and the original pathology recognition result indicates that the corresponding prostate lymph node does not include the target tissue portion, the first confidence is increased according to the similarity between the transition pathology feature image and the most similar pathology example image to obtain the target confidence. Including the target tissue, the first confidence is increased according to the similarity between the transition pathological feature image and the most similar pathological example image to obtain the target confidence; if the attribute corresponding to the target pathological image classification cluster is that the pathological example image does not include the target tissue part, and the original pathological recognition result indicates that the corresponding prostate lymph node includes the target tissue, the first confidence is reduced according to the similarity between the transition pathological feature image and the most similar pathological example image to obtain the target confidence; if the attribute corresponding to the target pathological image classification cluster is that the pathological example image does not include the target tissue part, and the original pathological recognition result indicates that the corresponding prostate lymph node includes the target tissue, the first confidence is reduced according to the similarity between the transition pathological feature image and the most similar pathological example image to obtain the target confidence. Among them, the specific way to increase the first confidence is various, and the similarity can be increased to the first confidence in a weighted manner to obtain the target confidence, or the equivalent value of the similarity can be determined by a certain equivalent formula, and the equivalent value is added to the first confidence to obtain the target confidence, etc., which is not specifically limited here. In addition, there are various ways to reduce the first confidence level. The similarity can be reduced to the first confidence level in a weighted manner to obtain the target confidence level. The equivalent value of the similarity can also be determined by a certain equivalent formula, and the equivalent value can be deducted from the first confidence level, etc., which are not specifically limited here.

[0055] Reference Figure 7 As shown, Figure 7 Shows Figure 6 A sub-step of a specific embodiment of step 602, in one embodiment, the preset pathology image library may include a pathology image classification label corresponding to each pathology image classification cluster, and a second confidence corresponding to each pathology example image, and step 602 may include the following steps.

[0056] Step 701: determining a target preset correction offset value corresponding to the target pathological image classification cluster from a plurality of preset correction offset values ​​according to the pathological image classification label of the target pathological image classification cluster; Step 702: Correct the first confidence level according to a plurality of second confidence levels corresponding to the target pathological image classification cluster and a target preset correction offset value to obtain a target confidence level.

[0057] In one embodiment, the pathological image classification label can indicate whether the target tissue part is included in the pathological image classification cluster. The second confidence is obtained by pathological identification based on the pathological example image, and the recognition confidence corresponding to the pathological image classification label. For example, the pathological example image is identified as a prostate lymph node that does not include the target tissue in the pathological identification, and the recognition confidence is 70%, and the pathological image classification cluster of the pathological example image is a prostate lymph node that includes the target tissue, then the second confidence corresponding to the pathological example image is 30%. In addition, multiple preset correction offset values ​​correspond to multiple pathological image classification clusters one by one, and the target preset correction offset value can be 0.1, 0.2, 0.3, 0.4, 0.5, etc. Those skilled in the art can determine the specific value of each preset correction offset value according to the actual situation, for example, according to the pathological image classification label of the target pathological image classification cluster, the specific positive and negative situation of the target preset correction offset value is determined, etc., which is not specifically limited here. In addition, multiple preset correction offset values ​​can be stored through a preset pathological image library, or can be stored in other ways, which is not specifically limited here.

[0058] In order to make pathology recognition have a certain generalization performance, certain recognition capabilities will be discarded when training related networks or executing architectures. In this case, pathology recognition will actually have a certain bias towards certain data, with better judgment capabilities for certain types of data and poorer judgment capabilities for certain types of data. For example, pathology recognition is more biased towards pathology images with larger target tissues on prostate lymph nodes, pathology images with color distribution domains similar to the color distribution domain of the training set, and other pathology images. These pathology images are easier to accurately recognize, while pathology recognition is less biased towards pathology images with smaller and scattered target tissues on prostate lymph nodes, pathology images with a large difference in color distribution domain from the color distribution domain of the training set, and other pathology images. These pathology images are more difficult to accurately recognize. This results in the fact that the recognition results of pathology recognition are actually unstable, resulting in the above-mentioned errors.

[0059] As for the target pathology image classification cluster, since the target pathology image classification cluster includes various forms of pathology example images, the difficulty of the pathology image corresponding to the target pathology image classification cluster in pathology recognition can be determined according to the second confidence of each pathology example image and the pathology image classification label corresponding to the target pathology image classification cluster. The lower the overall second confidence of the target pathology image classification cluster, the lower the recognition accuracy of the pathology image corresponding to the target pathology image classification cluster in pathology recognition, and the more prone to errors in the original mask image and the original pathology recognition results. To this end, the first confidence can be adjusted by the difficulty of the pathology image corresponding to the type of the target pathology image classification cluster in pathology recognition, so that the difficulty of the pathology image corresponding to the type of the target pathology image classification cluster in pathology recognition can be reduced to reduce the impact caused by the compromise of generalization performance in pathology recognition, and obtain a target confidence with higher reliability.

[0060] Specifically, first, a preset correction offset value can be set for various pathological image classification clusters to indicate the degree of offset of pathological images of the types corresponding to the various pathological image classification clusters during the pathological recognition process. Next, the difficulty of pathological images of the types corresponding to the target pathological image classification cluster in pathological recognition is determined according to multiple second confidences corresponding to the target pathological image classification cluster. The first confidence is corrected according to the preset correction offset value and the difficulty of pathological images of the types corresponding to the target pathological image classification cluster in pathological recognition to obtain the target confidence.

[0061] It should be noted that the recognition result corresponding to the obtained target confidence (that is, the target pathology recognition result) may be contrary to or the same as the original pathology recognition result. If a pathology recognition result needs to be output, a more accurate target pathology recognition result may be output, which is not specifically limited here.

[0062] In one embodiment, step 702 can be implemented as follows: first, the recognition difficulty is analyzed according to the multiple second confidences corresponding to the target pathological image classification cluster to obtain the comprehensive confidence corresponding to the target pathological image classification cluster, and then the comprehensive confidence and the target preset correction offset value are superimposed on the first confidence to correct the first confidence to obtain the target confidence. The recognition difficulty is quantified by the comprehensive confidence, so that the first confidence can more accurately correct the first confidence in the degree of difficulty of pathological images of the type corresponding to the target pathological image classification cluster in pathological recognition, thereby reducing the impact of pathological recognition due to the compromise of generalization performance and improving the stability of pathological recognition for pathological images of the type corresponding to the target pathological image classification cluster.

[0063] Reference Figure 8 As shown, Figure 8 Shows Figure 7A sub-step of a specific embodiment of step 702 in the embodiment, step 702 may include the following steps.

[0064] Step 801: determining a third confidence level and a fourth confidence level among a plurality of second confidence levels corresponding to a target pathological image classification cluster; Step 802: Correct the first confidence level according to the third confidence level, the fourth confidence level and the target preset correction offset value to obtain the target confidence level.

[0065] In one embodiment, the third confidence level and the fourth confidence level may be two random confidence levels among multiple second confidence levels corresponding to the target pathology image classification cluster, or may be two specific confidence levels among multiple second confidence levels corresponding to the target pathology image classification cluster, which is not specifically limited here.

[0066] In one embodiment, the target confidence level may be obtained by superimposing the difference between the third confidence level and the fourth confidence level and the target preset correction offset value onto the first confidence level.

[0067] In one embodiment, step 402 may obtain the target confidence by the following formula (1).

[0068] (1), In formula (1), Can represent target confidence; It can represent the first confidence level; A confidence set formed by multiple second confidences corresponding to the target classification cluster may be represented; It can indicate the preset reliability correction factor; A third confidence level can be expressed; It can represent the fourth confidence level; It can indicate the target preset correction offset value.

[0069] When the third confidence level and the fourth confidence level are respectively the minimum and maximum values ​​of multiple second confidence levels corresponding to the target pathology image classification cluster, the influence of the most difficult sample to correctly identify in the target pathology image classification cluster on the correction process of the first confidence level is introduced through the third confidence level, so that the correction process of the first confidence level can be performed through the gap between the transition pathology feature image and the pathology example image corresponding to the third confidence level, thereby improving the correction ability of the first confidence level, and further reducing the influence of the compromise of the generalization performance of pathology recognition on the first confidence level, so that the reliability of the target confidence level can be improved.

[0070] Reference Fig. 9 As shown, Fig. 9 Shows Figure 1 A sub-step of a specific embodiment of step 104 in the embodiment, in one embodiment, step 104 may include the following steps.

[0071] Step 901: determining a target pathology recognition result according to a target confidence level; Step 902: when the target pathology recognition result indicates that the prostate lymph node includes the target tissue, a plurality of target pixels are generated according to the target confidence and the target tissue mask portion; Step 903: Perform pixel replacement on the target tissue mask portion according to multiple target pixels, and output a target pathology mask image.

[0072] In one embodiment, under the premise that the original pathology mask image includes the target tissue mask part, when the target pathology recognition result indicates that the prostate lymph node includes the target tissue, it can be determined that the target tissue mask information in the original pathology mask image is correct. On this basis, multiple target pixels can be generated according to the target confidence and the target tissue mask part, and the target tissue mask part can be pixel-replaced, thereby improving the distinction between the target tissue mask part and other mask parts, so that medical staff can more easily see the target tissue in the prostate lymph node in the target pathology mask image.

[0073] In one embodiment, under the premise that the original pathology mask image includes the target tissue mask part, when the target pathology recognition result indicates that the prostate lymph node does not include the target tissue, pixel restoration is performed according to the target confidence and the target tissue mask part to delete the target tissue mask information in the original pathology mask image. The pixel restoration can be achieved by determining a restoration coefficient through the target confidence, and generating multiple restored pixels according to the restoration coefficient and the pixel values ​​of the pixels of the target tissue mask part (that is, the original pixels appearing below), etc., which is not specifically limited here.

[0074] In one embodiment, after step 903, the transition pathological feature image and the first confidence level may be stored in a preset pathological image library, thereby increasing the number of difficult samples in the target classification cluster, and further improving the accuracy of the target pathological mask image corresponding to the subsequent pathological image.

[0075] In one embodiment, step 902 can be implemented as follows: a modified target weight is determined according to the target confidence, the target tissue mask portion is convolved by a preset convolution kernel and the modified target weight, a target pixel corresponding to each pixel in the target tissue mask portion is generated, and the pixels of the target tissue mask portion are replaced by these target pixels. The modified target weight determined by the target confidence and the preset convolution kernel enable the target tissue mask portion to be enhanced in detail by the target confidence, thereby improving the masking effect of the target pathology mask image.

[0076] Reference Fig.10 As shown, Fig.10 Shows Fig. 9 A sub-step of a specific embodiment of step 902 in the above description, in one embodiment, the target tissue mask portion includes a plurality of original pixels, and step 902 may include the following steps.

[0077] Step 1001: for each original pixel, adjust the pixel value of the original pixel according to the target confidence level to obtain the target pixel value corresponding to the original pixel; Step 1002: Generate a plurality of target pixels according to a plurality of target pixel values.

[0078] In one embodiment, step 1001 can be expressed by the following formula (2).

[0079] (2), In formula (2), The pixel value of the target pixel can be represented; The pixel value that can represent the original pixel; It can represent the target confidence obtained in formula (1); It can represent a preset pixel correction coefficient, wherein the preset pixel correction coefficient can be 0.1, 0.2, 0.3, 0.4, 0.5, etc., and is not specifically limited; It can represent a preset pixel value threshold, wherein the preset pixel value threshold can be determined according to the attributes of the pixel value of the original pixel. When the pixel value of the original pixel has been subjected to interval conversion processing during storage, the preset pixel value threshold can correspond to 255; when the pixel value of the original pixel has not been subjected to interval conversion processing during storage, the preset pixel value threshold can correspond to 1. No specific limitation is made here.

[0080] By correcting the pixel value of the original pixel by using the target confidence, the preset pixel correction coefficient and the preset pixel value threshold, the pixel value of the target pixel can be quickly obtained, thereby quickly improving the generation efficiency of the target pixel; and, since the minimum value of multiple second confidences corresponding to the target pathology image classification cluster is introduced in the process of correcting the first confidence, the difference between the transition pathology feature image and the most difficult sample in the target pathology image classification cluster is improved, thereby improving the ability to correct the pixel value of the original pixel, and further making the target tissue mask information carried by the target pathology mask image more accurate and obvious.

[0081] In one embodiment, the pathological image is obtained by acquiring a pathological slice image of a prostate lymph node and performing a sliding window cropping process on the pathological slice image, wherein the pathological images obtained by performing a sliding window cropping process on the pathological slice image are multiple. After step 104, the histopathological recognition method of prostate lymph node metastasis may also include the following contents: performing pathological recognition on the next pathological image to output a target pathological mask image corresponding to the next pathological image, after obtaining the target pathological mask images corresponding to the multiple pathological images, performing graphic merging according to the target pathological mask images corresponding to the multiple pathological images to obtain a target pathological slice mask image corresponding to the pathological slice image, wherein the target pathological slice mask image includes a target tissue body mask portion of the prostate lymph node. In addition, the image merging may include merging and edge smoothing, thereby improving the visual effect of the target tissue body mask in the target pathological slice mask image. By performing a sliding window cropping process on the pathological slice image of the prostate lymph node, the feature quantity of the pathological slice image can be reduced, thereby reducing the difficulty of outputting the target pathological slice mask image, so that the histopathological recognition efficiency of prostate lymph node metastasis can be improved.

[0082] Reference Fig.11 As shown, the embodiment of the present application further discloses a histopathological identification device for prostate lymph node metastasis. The histopathological identification device 1100 for prostate lymph node metastasis can implement the histopathological identification method in the previous embodiment. The histopathological identification device 1100 includes: The original pathology recognition module 1101 is used to perform pathology recognition based on the pathology image of the prostate lymph node and obtain a first confidence level corresponding to the original pathology recognition result; The original pathology mask processing module 1102 is used to perform mask processing on the pathology image to obtain an original pathology mask image; A target confidence acquisition module 1103 is used to correct the first confidence according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence; The target pathology mask image output module 1104 is used to perform pixel correction on the target tissue mask part according to the target confidence when the original pathology mask image includes the target tissue mask part, and output the target pathology mask image.

[0083] It should be noted that, since the tissue pathology identification device 1100 of this embodiment can implement the tissue pathology identification method of the previous embodiment, the tissue pathology identification device 1100 of this embodiment and the tissue pathology identification method of the previous embodiment have the same technical principles and the same beneficial effects, and in order to avoid repetition of content, they will not be repeated here.

[0084] Reference Fig.12 As shown, the embodiment of the present application further discloses an electronic device, the electronic device 1200 includes: at least one processor 1201; At least one memory 1202, used to store at least one program; When at least one program is executed by at least one processor 1201 , the above-mentioned tissue pathology identification method is implemented.

[0085] The embodiment of the present application further discloses a computer-readable storage medium, which stores a computer program executable by a processor. When the computer program executable by the processor is executed by the processor, it is used to implement the above-mentioned tissue pathology identification method.

[0086] An embodiment of the present application also discloses a computer program product, including a computer program or computer instructions, wherein the computer program or computer instructions are stored in a computer-readable storage medium, a processor of an electronic device reads the computer program or computer instructions from the computer-readable storage medium, and the processor executes the computer program or computer instructions, so that the electronic device performs the above-mentioned tissue pathology identification method.

[0087] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0088] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0089] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0090] In the embodiments of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program that has a predetermined function and works together with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0091] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0092] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0093] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store program codes.

[0094] The step numbers in the above method embodiment are only provided for the convenience of explanation and description, and no limitation is imposed on the order of the steps. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

Claims

1. A histopathological identification method for prostate cancer lymph node metastasis, characterized in that: The following steps are involved: Performing pathological recognition according to the pathological image of the prostate lymph node to obtain a first confidence level corresponding to the original pathological recognition result; Performing mask processing on the pathological image to obtain an original pathological mask image; Correcting the first confidence level according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence level; When the original pathology mask image includes a target tissue mask portion, pixel correction is performed on the target tissue mask portion according to the target confidence level, and a target pathology mask image is output.

2. The method according to claim 1, characterized in that The plurality of pathological example images form a plurality of pathological image classification clusters in the preset pathological image library; The performing pathology recognition according to the pathology image to obtain a first confidence level corresponding to the original pathology recognition result includes: Extracting features from the pathological image to obtain a transitional pathological feature image; The step of correcting the first confidence level according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence level includes: Clustering the transition pathology feature images according to the plurality of pathology image classification clusters, and determining a target pathology image classification cluster from the plurality of pathology image classification clusters, wherein the target pathology image classification cluster includes the plurality of pathology example images; The first confidence is corrected according to a plurality of the pathological example images in the target pathological image classification cluster to obtain the target confidence.

3. The method according to claim 2, characterized in that The preset pathological image library includes a pathological image classification label corresponding to each of the pathological image classification clusters, and a second confidence level corresponding to each of the pathological example images, wherein the second confidence level is obtained by performing pathological recognition on the pathological example images and is a recognition confidence level corresponding to the pathological image classification label; The step of correcting the first confidence level according to the plurality of pathological example images in the target pathological image classification cluster to obtain the target confidence level includes: According to the pathological image classification label of the target pathological image classification cluster, determining a target preset correction offset value corresponding to the target pathological image classification cluster from a plurality of preset correction offset values, wherein the plurality of preset correction offset values ​​correspond one-to-one to the plurality of pathological image classification clusters; The first confidence is corrected according to the plurality of second confidences corresponding to the target pathological image classification cluster and the target preset correction offset value to obtain the target confidence.

4. The method according to claim 3, characterized in that The step of correcting the first confidence level according to the plurality of second confidence levels corresponding to the target pathological image classification cluster and the target preset correction offset value to obtain the target confidence level includes: Determining a third confidence level and a fourth confidence level among the plurality of second confidence levels corresponding to the target pathological image classification cluster; The first confidence level is corrected according to the third confidence level, the fourth confidence level and the target preset correction offset value to obtain the target confidence level.

5. The method according to claim 1, characterized in that: The pixel correction of the target tissue mask part according to the target confidence level and outputting a target pathology mask image comprises: determining a target pathology recognition result according to the target confidence; When the target pathology recognition result indicates that the prostate lymph node includes the target tissue, a plurality of target pixels are generated according to the target confidence and the target tissue mask portion; The target tissue mask portion is pixel-replaced according to the plurality of target pixels, and the target pathology mask image is output.

6. The method according to claim 5, characterized in that The target tissue mask portion includes a plurality of original pixels; The generating a plurality of target pixels according to the target confidence and the target tissue mask portion comprises: For each of the original pixels, adjusting the pixel value of the original pixel according to the target confidence level to obtain a target pixel value corresponding to the original pixel; A plurality of target pixels are correspondingly generated according to the plurality of target pixel values.

7. The method according to claim 1, characterized in that The pathological image is obtained by acquiring a pathological slice image of a prostate lymph node and performing a sliding window cropping process on the pathological slice image, wherein the pathological image obtained by performing a sliding window cropping process on the pathological slice image is a plurality of images; The method further comprises: Performing pathology recognition on the next pathology image to output the target pathology mask image corresponding to the next pathology image; After obtaining the target pathology mask images corresponding to the multiple pathology images, graphics are merged according to the target pathology mask images corresponding to the multiple pathology images to obtain the target pathology slice mask image corresponding to the pathology slice image, wherein the target pathology slice mask image includes the target tissue main body mask part of the prostate lymph node.

8. A histopathological identification device for prostate cancer lymph node metastasis, characterized in that: include: The original pathology recognition module is used to perform pathology recognition based on the pathology image of the prostate lymph node and obtain a first confidence level corresponding to the original pathology recognition result; An original pathology mask processing module, used for performing mask processing on the pathology image to obtain an original pathology mask image; A target confidence acquisition module, used for correcting the first confidence according to a plurality of pathological example images in a preset pathological image library to obtain a target confidence; The target pathology mask image output module is used to perform pixel correction on the target tissue mask part according to the target confidence when the original pathology mask image includes the target tissue mask part, and output the target pathology mask image.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program executable by a processor is stored therein, and when the computer program executable by the processor is executed by the processor, it is used to implement the method according to any one of claims 1 to 7.

Images