Chest Ray Disease Identification Based On Deep Learning Robust Method

Thorax Diseases has always been one of the main culprits affecting human life and health,so there is an urgent need for a precise classification and detection system of medical images for thorax diseases,which can help the radiologist to improve their efficiency and reduce their pressure.In recent years,with the rise of deep learning,more researchers are engaged in thorax diseases recognition and detection tasks based on deep learning,and have achieved a lot of results.However,these studies do not solve the diversity and correlation problems between different thorax diseases.In addition,the category imbalance,the limited data label information and the label noise are also the difficulties that will affect the final performance.To solve these problems,we design an end-to-end thorax disease recognition and detection system based on deep learning,with the latest SE and Spatial pooling module,which can reuse information from through space and channel to solve the problem of thorax disease diversity and correlation.In addition,we propose a batch adaptive weight method,which uses multi-batch samples to calculate category weights,so that the model can effectively solve the category imbalance problem in training while avoiding the model falling into the noise label of the dataset.Our model learns with the largest and most authoritative public dataset——ChestX-rayl4,to classify and locate total 14 thorax diseases.We compare with the existing methods on classification and identification of thorax diseases based on deep learning.The experimental results prove the superiority of the model algorithm.The final performance index reaches 0.8279.At the same time,we set up a self-contrast experiment to prove the effectiveness of all methods in this paper.Moreover,we set up the experiment of artificial noise,which proves the anti-noise performance of the batch adaptive weight proposed in this paper.