Healthcare Research at MSR India

At Microsoft Research Lab India, we conduct a variety of healthcare related research, including smartphone-based low-cost diagnostics, generative AI chatbots to support the healthcare ecosystem, and promote mental well-being of employees.

Low-cost diagnostics

Healthcare is not accessible to a huge population across the globe. There are a variety of reasons for that including skewed doctor-to-patient ratio, unskilled to semi-skilled healthcare workers, long waiting and long commute to see a doctor which also results in loss of income for daily wage earners. Our aim is to democratize healthcare. And our vision is to develop low-cost smartphone-based diagnostic solutions with AI assistant embedded in them, to enable community health workers, teachers, primary clinicians, and even Swiggy/Amazon delivery personnel, to perform preliminary screening of certain diseases with minimal training. With that vision in mind, over the past four years, we have been working towards developing a variety of diagnostics tool—detecting keratoconus using a smartphone-based corneal topographer (SmartKC), estimating refractive errors (Auto-Retinoscopy), computing the dryness level of the eye, classifying crackle and wheeze lung sound (RespireNet (opens in new tab)), and estimating height (opens in new tab) of children for malnutrition prediction—all using smartphone. These works has been done in close collaboration with hospitals like Sankara Eye Hospital and NGO’s like WeltHungerHilfe.

HealthBots

Patients and their caregivers require timely, trustworthy, detailed, and accurate information about their condition and treatment. Access to such information can help reduce anxiety and improve preparedness before, during, and after treatment. To address this need, we designed and developed a family of AI-powered HealthBots that leverage state-of-the-art generative AI models grounded in clinician-provided knowledge bases. These systems, including CataractBot (opens in new tab), OncoBot, and DRBot, assist patients and caregivers by answering questions related to pre-treatment preparation, treatment procedures, recovery, and post-treatment care. A key aspect of these systems is the expert-in-the-loop component, wherein healthcare professionals and patient coordinators review bot responses and provide feedback, enabling continuous improvement in answer quality over time. The bots are designed to be multimodal and multilingual. Learn more about the HealthBots project.

Building on this work, we have extended the HealthBot paradigm beyond patients to support frontline healthcare workers through ASHABot (opens in new tab), an AI assistant designed to help Accredited Social Health Activists (ASHAs) access reliable health information and guidance in their day-to-day work.

Radiology

This research investigates how AI assistants can be integrated into radiology workflows in a reliable and clinically meaningful manner. As part of this effort, we have developed models such as RadPhi-3 and Rad-Phi4-Vision-CXR.