Research On Auxiliary Diagnosis Of Breast Cancer Based On Deep Learnin

Breast cancer is the most common cancer in women,with the highest incidence and mortality rates among all types of cancers.Early detection,early diagnosis and early treatment can improve the cure rate of patients.Analyzing pathological images is the key for doctors to determine whether a patient has developed cancer,the specific type of cancer,and the severity of the cancer.The use of computer image processing technology can assist doctors in pathological diagnosis and effectively improve their work efficiency.Medical image processing based on deep learning technology has become a research focus in the fields of medicine and computer science in recent years.Under this research background,this paper conducts indepth research on breast cancer cell nucleus segmentation and tissue region segmentation based on deep learning,aiming to improve the accuracy of breast cancer diagnosis by improving the model segmentation ability.This thesis starts on the basis of existing domestic and foreign related researches,and the main works are as follows.(1)This thesis proposes a Multifunctional Aggregation Network(MA-Net)for addressing the inaccurate cell nucleus segmentation of breast cancer cells caused by cell adhesion and small target recognition issues.The model is based on U-Net and incorporates fusion modules to increase feature extraction and reuse,as well as context extractor modules to recognize targets of different sizes.To address the problem of cell adhesion,an attention gate module is added to the skip-connection structure to enable the model to pay more attention to segmentation details.Furthermore,residual modules are used to replace the U-Net network convolution module to alleviate gradient explosion and vanishing problems and to form the MA-Net model.The improved model achieved mean intersection-overunion(MIoU)scores of 0.7826 and 0.8047 on the MoNuSeg and TNBC datasets,respectively,which are 6%and 5%higher than those of U-Net.The experimental results demonstrate that MA-Net has good generalization and comprehensive segmentation capabilities..(2)An efficient preprocessing approach was proposed for Whole Slide Images(WSI).HSV threshold segmentation and sliding window extraction were used to extract the foreground region of the thumbnail image,and the corresponding coordinates were mapped to the high-resolution image for slicing to obtain small-sized slices for model training and prediction.The MA-Net model was used to perform tissue region segmentation experiments on the Camelyon16 and Camelyon17 breast cancer WSI datasets,with a precision scores of 0.9874 achieved,achieving accurate segmentation of the lesion tissue region.(3)Based on the research results of this thesis on breast cancer pathological images,a breast cancer disease medical auxiliary diagnosis system prototype was designed using PyQt5.By predicting the lesion proportion of pathological images,the system efficiently assists doctors in the preliminary screening and cancer grading diagnosis of pathological images.