Analysis and Optimum Classification of Thoracic Disease in Chest X-Rays using ResNet CNN

Abstract

Computer vision and image diagnosis have seen tremendous improvement in the last decade. This is due to the insurmountable use cases created in health care and other fields using artificial neural networks. Heart and lung failure deaths consist of more than 1,000,000 occurring every year due to poor health care system as per the government report. Now with computer vision, we can analyze any type of Chest X-ray without a doctor’s consultation which can save millions of lives. Image diagnosis using CNNs is very cost-efficient and reliable. The main reason for this technology not being in use today is the difficulty in predicting lung infections as the Chest X-Rays contain disease tissues with lower contrast spots along with the other tissues. Also, there is an N number of chest tissue diseases each with a unique pattern that causes lower accuracy. The existing solutions still have a lower accuracy rate. We have used the ResNet deep neural network model with specific techniques implemented on top of it to predict thoracic disease prediction. This model has considerable advantages like no vanishing gradient problem over others and thus leading to better accuracy.