Research On Chest X-ray Lesion Region Recognition Method Based On Convolution Neural Network

Thoracic diseases are a common condition that considerably impacts human health problems worldwide.Chest X-ray is one of the most common medical imaging technologies for diagnosing thoracic diseases.The automatic recognition of thoracic diseases and the localization of the lesion area have important research value and practical significance,because only rely on the radiologists to analyze the chest X-ray images at present.This thesis takes chest X-ray images as the research object,and deeply explores the multi-scale pathological characteristics and lesion location of thoracic diseases in chest X-ray images.The main innovative work includes:(1)There are significant scale changes in pathological abnormalities of different thoracic diseases in chest X-ray images,and the standard 3×3 convolution unable adapt to the multiscale features of different diseases.The convolution neural network is difficult to comprehensively learn the complex pathological abnormal features of thoracic diseases due to each chest X-ray image has similar organs and tissues.To solve the above problems,this thesis proposes a residual network based on pyramid convolution and shuffle attention(PCSANet).Compared with standard 3×3 convolution,pyramid convolution is used to extract more multi-scale features of pathological abnormalities;shuffle attention enables PCSANet to focus on more pathological abnormal features.Extensive experiments on the Chest X-ray14 and COVIDx datasets show that PCSANet has better performance(average AUC = 0.825)compared to state-of-the-art methods.(2)Most existing approaches typically employ a global learning strategy and use convolution neural network with small convolutional kernels for thoracic disease classification.However,irrelevant noisy regions may affect the global learning strategy;small convolutional kernels can only capture fewer discriminant features.To address the above problems,we construct a multi-feature fusion neural network(MFNet),which can fully use the global and local features.Specifically,the global features are first generated by the global branch.The local features are generated by multiplying the global feature and the lung-heart region mask identified by the Lung-heart Region Generator(LHRG).At last,the fusion branch integrates the global and local features to complement the lost discriminative feature of the global branch and the local branch,thus enabling a better feature presentation for thoracic disease classification.Extensive experiments on the Chest X-ray14 dataset demonstrate that the MFNet achieves superior performance(average AUC=0.844)compared to state-of-the-art methods.