Chest X-ray Aided Diagnosis Algorithm Based On Deep Learning

Chest X-ray is the most common medical imaging test in the world and is essential for the diagnosis of a variety of chest diseases,including tuberculosis and lung cancer,with more than 2 billion uses per year.The method of relying entirely on radiologists for chest X-ray diagnosis is not only inefficient,but also the results of the diagnosis are related to the professional ability level of the doctor.It may even be misdiagnosed due to fatigue of the reading,and the accuracy of the diagnosis is difficult to be effectively guaranteed.Therefore,it is very important to study a chest x-ray aided diagnosis algorithm.The traditional chest x-ray aided diagnosis algorithm relies on the algorithm designer to manually perform feature extraction based on specific disease characteristics,such as the shape,size,and gray scale of the lesion,and then analyze the extracted features to achieve auxiliary diagnosis of the disease,but This method must design different feature extraction methods for different disease characteristics,which is time consuming and laborious.Deep learning has achieved great success in the field of image recognition with its excellent automatic feature extraction ability.The use of deep learning to assist diagnosis of chest disease has become a hot research topic.Therefore,based on deep learning,this paper proposes a chest-assisted diagnosis algorithm based on multi-network feature fusion.The algorithm consists of three parts.The first part is feature extraction.The algorithm uses two different neural networks,ResNet and DenseNet,to extract features.The two networks are independent of each other.Feature extraction,and in order to speed up the training speed of the algorithm,the parameters used in the network for feature extraction are not trained from the beginning,but are initialized using the pre-trained parameters in the ImageNet dataset,and then the chest image is used.The second part is the feature fusion,which combines the features extracted by the two neural networks.When the fusion,the two features are connected by channels,and the information in the two features is retained to the maximum,and then the average pooling operation is used.The connected features are averaged over each channel to obtain a fused feature.The third part is a multi-label classifier.Each chestpiece may contain multiple diseases.Therefore,chest-slice-assisted diagnosis is a multi-label classification task.The algorithm constructs a multi-label classification based on a fully connected layer with Sigmoid function as the activation function.A multi-label classification of the fusion features of the chest radiograph to achieve an auxiliary diagnosis of chest disease.This paper conducted an experiment on the ChestX-ray14 dataset published by the National Institutes of Health(NIH)Clinical Center,classifying and identifying 14 chest X-ray diseases in the ChestX-ray14 dataset,and comparing them with other similar algorithms.The evaluation index AUC has achieved a certain improvement.