Research And Application Of Immunohistochemical Image Classification Method For Color Abnormality

Pathological examination is the gold standard for clinical diagnosis of many diseases.Immunohistochemistry has been widely used in pathological evaluation to provide reliable evaluation results and effective treatment options for patients.With the development of digital physical,immunohistochemical tissue sections can be digitized to obtain immunohistochemical images,which are easier to save and observe.However,in the face of massive immunohistochemical images,pathologists adopt the method of visual diagnosis,which not only has a heavy burden of reading but also is prone to subjective differences.In recent years,with the rapid development of deep learning in disease-assisted diagnosis,its application in immunohistochemical image analysis can greatly simplify the reading process of pathologists.However,the existing immunohistochemical image classification methods based on deep learning ignore the consideration of limited immunohistochemical image resources,poor model interpretability and weak ability to extract immunohistochemical image features.In view of the above problems,the main research work of this paper is as follows:(1)This paper proposes an improved feature importance network to complete the classification task of immunohistochemical images.Without additional expansion of the immunohistochemical image dataset,by optimizing the data enhancement method in the comparative learning model and promoting the interactive learning of the online network and the target network,the online network learning can effectively generalize the immunohistochemical images and provide prior knowledge for FIN.At the same time,the attention mechanism in the improved FIN improves the interpretability of the classification model,and effectively guides the model to focus on the region of interest in the immunohistochemical image,which is consistent with the region of interest of the pathologist.The experimental results show that the precision,Recall and Fl-score indexes of the improved FIN method are improved on the CD34 and DeepLIIF datasets,and the Fl-score index is the highest.Compared with the FIN method,it is improved by 5.42%on the CD34 dataset and 13.21%on DeepLIIF.(2)This paper proposes an immunohistochemical image classification method that incorporates the CD perception branch.This method first uses the color deconvolution algorithm to extract the positive staining area of the immunohistochemical image and reconstruct the image to obtain the color anomaly map,and then uses the feature extraction network to obtain the intensity and area features of the positive staining in the color anomaly map to obtain the index information consistent with the pathologist’s attention.Finally,the index features are fused with the immunohistochemical image features in a way that introduces external knowledge,so that the model can effectively focus on the index features of the image while paying attention to the original immunohistochemical image features.The experimental results show that the method of integrating CD perception branch into FIN and improved FIN improves the Precision,Recall and F1-score indicators on CD34 and DeepLIIF datasets,and the F1-score indicator has the highest improvement.Compared with the FIN method,the method of integrating the CD perception branch into the FIN is improved by 7.22%on the CD34 dataset and 10.05%on the DeepLIIF.The method of integrating the CD perception branch into the improved FIN is improved by 9.22%on the CD34 dataset and 14.72%on the DeepLIIF.(3)In order to realize the classification operation of immunohistochemical images and the visualization of classification results,this paper designs and implements an immunohistochemical image scoring system.By embedding the classification model based on the research content of this paper into the system,the classification task of immunohistochemical images and the display of classification results can be realized.In addition,the construction of immunohistochemical image data sets and some imageprocessing operations can also be realized in the system.