The project was started by Dr. Geetha Manjunath of Niramai Health and Prof. Chiranjib Bhattacharyya of IISc, in the first wave of Pandemic to address the issue of Accessibility of Healthcare in the context of COVID-19. The key premise behind the project was X-rays are far more ubiquitous than RTPCR tests and a diagnostic tool based on X-ray images could be an important proxy for RTPCR. To ensure accessibility a tool was developed on Whatsapp platform for easy communication with practicing Doctors .
The technical challenge was to develop predictive models for understanding Whatsapp based X-ray images which are not only accurate but also provide some explanation with very few instances of COVID-19 positive cases. The model is based on an unique Transfer Learning framework that leverages easily available X-ray images of lungs, not necessarily COVID positive, to learn useful features which have high predictive power. A confidence score, guided by the infected areas of lungs, along with automatic annotation provides some insights into the final prediction. The system outputs a prediction, localizes the infected parts, and creates a report which gives a confidence score, all within a few minutes.
The system was piloted in Karnataka as COVID SWIFT ( link for description of the working system) over the last 10 months. Buoyed by the success of the pilot, in collaboration with Mr. Umakant Soni of ARTPARK, the tool have now been launched as X-ray Setu, a free-to-use AI-driven platform to aid doctors for early-Covid interventions over WhatsApp. For more information see Xray Setu Website.