Automatic Recognition And Classification Of Breast Tumor Pathological Image Based On Deep Learning

Pathological diagnosis,as the golden standard of cancer diagnosis,plays an increasingly important role in clinical medicine.However,with the development of pathological image acquisition technology,the workload of pathologists is increasing.An accurate and stable algorithm is needed to assist doctors to efficiently perform pathological diagnosis.The emergence of machine learning represented by deep learning provides a new way.Application of deep learning in pathology faces many issues.Based on basic principles of artificial neural network and the characteristics of pathological images,a deep learning-based automatic recognition for pathological images is proposed,which can improve the efficiency of doctors and enhance the objectivity of diagnosis.The main works are shown as follows:(1)Aiming at the shortage of label data in pathological field and the problem of over-fitting in deep learning,an automatic cell extraction algorithm based on image segmentation is proposed.Under the guidance of doctors,labels are added to construct the database.(2)Aiming at the problem that complex artificial neural network model is highly dependent on hardware and incompatible with medical equipment,a six-layer convolutional neural network model is proposed for automatic identification of cancer cells.The learning rate of periodic oscillation is used to train the model in order to obtain better convergence results through less iteration.Pathological experiments verify the effectiveness of the algorithm.(3)Aiming at the imbalance of tumor cell data in pathological images and the low recognition accuracy of minority classes in image recognition network,a sample equalization algorithm based on generative adversarial networks is proposed.Two types of generation networks,cGAN and cDCGAN are established by studying different generative adversarial models to accomplish sample equalization.The experimental results verify the effectiveness and robustness of the sample equalization algorithm.(4)To solve the problem of ambiguous criteria and high subjectivity in pathological diagnosis of breast tumors,a quantitative analysis method based on pixel classification and cell classification of pathological images is proposed in order to provide assistance for doctor's diagnosis through intuitive probability maps and objective cell count.