Analysis And Research On The Assistant Pathological Diagnosis Of Breast Cancer Based On Deep Learning

Breast cancer,as the most common cancer at present,seriously endangers human health.As the "gold standard" for breast cancer diagnosis,histopathological examination usually relies on experienced pathologists.In recent years,artificial intelligence technology has been applied to digital pathological images to help doctors complete breast tumor recognition work,mainly focusing on the classification of pathological image patches of breast tissue and the segmentation of cell nuclei.Research on breast cancer diagnosis based on the analysis of histopathological whole slide images is not mature enough.On this basis,this paper analyzes and judges whether the whole slide images of sentinel lymph nodes of breast cancer patients contain cancer cell metastases,in order to assist the pathological diagnosis of breast cancer.It mainly includes the following three parts of work:(1)Achieved the pretreatment of histopathological whole slide images.According to the characteristics of histopathological slide images,the threshold-based Otsu method was used to extract the tissue area,and randomly sampled to obtain image patches and labels from the normal tissue area and the lesion area.The contrast-limited adaptive histogram equalization method was used to enhance the extracted image patches,the method effectively highlights the morphological characteristics of the cells.(2)Achieved the segmentation of histopathological image patches.Combining the residual learning in the Res Net model and the U-Net model’s advantages of fusing deep and shallow features,using the residual structure instead of the traditional convolution structure,the R＿Unet network model was proposed.It has been verified that the R＿Unet model converges faster,and the segmentation performance of image patches is better than the U-Net model,the accuracy and Dice value reached 92.6% and 87.8%,at the same time,the amount of model parameters was reduced by 73.3%.(3)Achieved the classification of histopathological whole slide images.By extracting the relevant features of the probabilistic heatmap of histopathological images,a feature dataset of the lesion area was constructed.Based on the Stacking method,three different classifiers were combined to build a new ensemble model.It has been verified that the new model is better than the single classifier in accuracy and other indicators,the recall rate and AUC reached 96.9% and 95.0%,so that histopathological whole slide images can be classified more accurately and the missed detection rate of malignant images can be reduced.