Research On Classification And Prediction Of Adenocarcinoma Pathology Images Based On Deep Learnin

Adenocarcinoma is a life-threatening malignant tumor.Breast cancer,prostate cancer and lung cancer are all types of adenocarcinoma,with a high incidence rate.Pathological analysis of needle biopsy is one of the methods for diagnosing the severity of adenocarcinoma.Traditional pathological analysis methods rely on manual feature extraction,making classification methods based on machine learning less generalized.Therefore,the classification method of adenocarcinoma pathological images based on deep learning has received more attention from scholars.Based on this,there are the following problems in the research of adenocarcinoma pathological image classification:(1)There is a lack of a large publicly available dataset of pathological images of adenocarcinoma,and some datasets still have imbalanced data.The research focus is mostly on simple binary classification of benign and malignant biopsy pathological images,neglecting the detailed classification of various subtypes;(2)Some puncture biopsies did not obtain the central tissue of the tumor,and the Gleason Score(GS)of the pathological images may escalate after radical resection.In response to the above issues,this article conducted in-depth research in the following two aspects.(1)This paper proposes a classification model DAFLNet for breast cancer pathological images based on Densely Connected ConvolutionalNetworks(DenseNet),Attention mechanism and Focal Loss(FL)functions.The experimental results show that the accuracy of DAFLNet model in the classification of benign and malignant breast cancer pathological images is 99.1%,and the accuracy of DAFLNet model in the classification of eight breast subtypes is 95.5%.This shows that the accuracy of DAFLNet model in the classification of breast cancer pathological images is high in the case of uneven data.(2)A prediction model MATNet based on multi model feature fusion,attention mechanism,and transfer learning is proposed.This model is used to predict the pathological images of prostate cancer biopsy,and to conduct in-depth research on whether the Gleason Score is upgraded after radical resection.The experimental results show that the MATNet model performs well in predicting the GS6 upgrade task of prostate cancer pathological images,with the accuracy rate of 85.5%,the precision of 82.1%,the specificity of 85.4%,and the sensitivity of 85.7%.The research in this paper effectively solves the shortcomings of the current research on classification of pathological images of breast cancer and the research on the upgrade prediction of Gleason Score of pathological images of prostate cancer.The research on pathological image classification of breast cancer based on DAFLNet model not only completes the recognition of benign and malignant breast,but also realizes the recognition of various subtypes.Therefore,DAFLNet model can better provide clinical diagnosis reference for oncologists;Based on the MATNet model,the Gleason Score upgrade prediction study realizes the prediction of pathological images of GS6,providing strong support for pathologists to diagnose the degree of differentiation of prostate cancer.To sum up,the classification of breast cancer and the prediction of prostate cancer have important clinical significance,which is helpful for doctors to diagnose the condition of adenocarcinoma.