Multi thoracic disease classifier using Convolutional Neural Networks

Abstract

India faces acute shortage of radiologists. As per NCBI (National Center for Biotechnology Information), USA, India has one radiologist per 1,00,000 people. In past two years we have seen an unprecedented COVID-19 pandemic which has posed a huge burden on our health care infrastructure and health care professionals. The rural parts are hit worst struggling to provide lifesaving health care access causing millions of Indians to lose their lives. In this regard our paper focuses on developing an Artificial Intelligence (AI) based web application which may reduce the burden on healthcare professionals and help in timely diagnosis of chest x-ray findings without delays and also with precision. This will help to treat patients with utmost care, can avoid unnecessary surgeries and save lives. In the recent years AI empowered systems have proven to be dominant in all domains. AI which encompasses all the industries has been proven to be vital in healthcare by helping healthcare professionals in taking decisions and also in diagnosis and detection of several critical ailments like cancer and others. In this paper we have leveraged the transfer learning as benchmark to obtain the models for our task of chest image classification. We have run the experiment through the various standard models available retaining the identical experimental conditions and did the comparative analysis to evaluate them and to pick the best one among them. The results achieved show that Densenet-169 provided the best results with 95.56 percentage validation accuracy during model training which has been used for making predictions in the web application.