Web Based Application to Detect Tuberculosis using CXR

Abstract

Tuberculosis is a very infectious respiratory disease and is currently the leading cause of mortality worldwide, ranking higher than both malaria and HIV/AIDS. As a result, it is vital to promptly diagnose TB to limit its transmission, enhance preventative measures, and reduce the mortality rate associated with the disease. Various procedures and tools have been employed to diagnose TB early, practically all of which needed a visit to the doctor and were not available to the public. This work presents an automated and accurate approach for diagnosing TB that may be used by the general population and does not require special imaging equipment or conditions. An application will be developed for the detection of TB using CXRs and deep learning techniques. The application will use a convolutional neural network (CNN) to classify CXRs as normal or indicative of TB. The CNN will be trained on dataset of annotated CXRs to learn the relevant features for TB detection.