Around 500,000 people die every year in India due to TB. Between 2006 and 2014, Indian economy suffered a loss of 340 Billion USD due to TB. Drug resistant strains in India are worse and much more prevalent. The government of India has come up with a new TB policy in 2017 (freely viewable here):
https://tbcindia.gov.in/WriteReadData/NSP%20Draft%2020.02.2017%201.pdf which aims to have a 4 pillar DTBP policy of which the first pillar D (or diagnose) aims to have a high sensitivity diagnosis. How can AI help you ask ? By acting as a diagnosis agent as we try to show in our latest work. Drug responsive TB is generally detected using Sputum test. At a diagnosis center, the work to perform a Sputum test is twofold: 1. Zoom to a proper level, 2. Detect and locate any bacillus clusters (bacteria that cause TB). In this work, we try to automate task 2 using a cascade of Neural Networks as shown in the figure. With a larger dataset, the technology can be deployed at scale to partially automate TB diagnosis, even better can be extended to full automation if an equivalent dataset of zoom levels is made available.