Automatic Analysis Method Of Pathological Diagnosis Based On Deep Learning

Medical Images are important basis for the pathology diagnosis,and segmentation and classification are two key problems for medical image processing.For segmentation,doctors usually needs to labelling lesion area or different organs and tissue according to medical images.For classification,doctors diagnose weather the patient is infected and decide treatment programs according to medical images.However,analysis of the med-ical images manually takes a lot of time and effort,leading to low diagnosis efficiency.Deep learning method have the advantages of speed,accuracy and repeatability,which can significantly relieve the burden on doctors.Thereby,deep learning can free doctors from low-level image analyzing tasks,which highly fits current medical requirements.So it is inevitable to achieve automatically pathological diagnosis by deep learning technol-ogy.This paper aims at the automatically analysis technology of pathological diagnosis,and conduct researches on segmentation and classification technology of pathological di-agnosis based on deep learning.For segmentation problems of pathological diagnosis,this paper proposed a novel unsupervised domain adaptation framework for medical image segmentation.The pro-posed framework synergizes both pixel space and entropy space for domain alignment.In the pixel space,we introduce the attention mechanism into Cycle GAN,and enhance the semantic and geometric consistency of the target during the image style transformation.In entropy space,we utilize entropy minimization principle to force consistent image seg-mentation between well-annotated source domain and non-annotated target domain.The aligned ensemble of two representation spaces enables a well-trained segmentation model to effectively transfer from source domain to target domain.The experimental results demonstrate the effectiveness of the proposed framework.For the task of segmentation from cross-modality medical images,our proposed framework achieves state-of-the-art performance,with some specific metric even superior to those of supervised methods.For classification problems of pathological diagnosis,this paper proposed automatic diagnostic system for vaginal microflora pathological images,named Res Lab v1 model,which based on Deep Lab model and utilize dilated convolution to enlarge receptive field while avoiding loss of spatial resolution? introducing residual block to avoid gradient van-ishing and explosion of deep network.We use loss function force the model focusing on training samples that difficult to classify for unbalanced data.Simultaneously,using data augmentation to expand the small sample.According to the feature of vaginal microflora pathological image,we proposed optimization for the system structure from three aspects:depth of network,receptive field and attention mechanism.The ablation experiment is carried out to determine the best network architecture.The accuracy of Res Lab model exceeds VGG,Goog Le Net and Res Net.