Your smartphone can now screen for tuberculosis and diabetes — no blood test, no X-ray needed

A new AI-powered application developed by HealthAI Labs uses your phone's camera and microphone to detect early signs of two of India's biggest silent killers. India carries a burden that most people don't talk about at dinner tables. Over 2.69 million Indians developed tuberculosis last year. Another 101 million are living with diabetes — and more than half of them don't know it yet. These are not rare diseases. They are everywhere, in every city, every village, every income bracket. And the tragedy is that both conditions are largely treatable when caught early. The problem isn't treatment. It's detection. Getting a TB test in rural India means a sputum sample, a lab, a wait. Getting a diabetes diagnosis means a blood draw, a lab, a doctor visit. For hundreds of millions of people — especially those in semi-urban and rural areas — that chain of steps never happens. The disease progresses silently. At HealthAI Labs, we asked a simple question: what if the screening tool was already in your pocket?

How does screening TB from a cough actually work?

When someone has pulmonary tuberculosis, the infection changes the structure of their airways. Inflammation, mucus, and tissue damage alter how air moves through the lungs during a cough. These changes produce a distinctive acoustic signature — a pattern in the sound of the cough that is measurably different from a healthy cough, a COVID cough, or a bronchitis cough. Researchers at Stellenbosch University, Google Health, and institutions across India, Zambia, the Philippines, and South Africa have spent the last decade building AI models that can read these acoustic patterns. The Google HeAR model was trained on over 313 million audio samples. A dataset called CODA-TB contains over 733,000 cough recordings from seven countries. The best AI classifiers in this space can identify TB-pattern coughs with an accuracy measure (AUROC) of up to 0.94 — meaning they are right 94% of the time in research settings. The HealthAI CoughScan module asks you to cough three times into your phone's microphone. In under 15 seconds it analyses the frequency, duration, wetness, and pattern of your cough and generates a TB risk assessment. It also checks for COVID-19 patterns, bronchitis, whooping cough, and asthma signatures — all from the same three coughs.

And diabetes — from a camera?

Diabetes silently damages blood vessels over years before any obvious symptoms appear. But those vascular changes are visible — if you know where to look. Three places in particular are accessible to a smartphone camera: the fingertip, the nail bed, and the tongue. When you hold your fingertip over the rear camera with the torch on, the camera picks up the tiny fluctuations in blood volume with each heartbeat — a technique called remote photoplethysmography (rPPG). In a healthy person, the pulse waveform has a characteristic shape with a secondary bump called the dicrotic notch, caused by the elasticity of the arterial walls. In people with diabetes-related arterial stiffness, this notch flattens or disappears entirely. A 2023 study published in IEEE Access demonstrated that a lightweight AI model trained on single pulse waveforms could detect type 2 diabetes with an AUROC of 0.755 — meaningful accuracy for a contactless, zero-cost test. The nail bed tells a parallel story. Poor peripheral circulation — a hallmark of uncontrolled diabetes — shows up as pallor, colour changes, and nail thickening that an AI vision model can assess from a photograph. The tongue, too, carries diagnostic information: diabetics are significantly more prone to white coating (oral candidiasis from high blood sugar), dryness, and specific colour changes that researchers at IIT Madras and IIT Bombay have been studying using deep learning models. The HealthAI GlycoCam module combines all four signals — PPG waveform, nail image, tongue image, and a brief risk questionnaire — into a single composite diabetes risk score. No needle. No lab. No waiting room.

This is a screening tool, not a diagnosis

We want to be completely transparent about what HealthAI is and what it isn't. It is not a replacement for a doctor, a blood test, or a sputum sample. A positive screen on HealthAI means you should seek clinical evaluation — not that you definitively have TB or diabetes. The science supporting these methods is real and published in peer-reviewed journals, but the HealthAI application itself is currently a screening and risk-awareness tool awaiting formal clinical validation in Indian populations. Think of it the way you think of a home blood pressure monitor. It doesn't replace your cardiologist. But it might be the thing that tells you to go see one.

What comes next

We are currently designing a formal pilot study in collaboration with a medical college in India to validate our screening modules against gold-standard reference tests — GeneXpert for TB and HbA1c blood testing for diabetes. We are looking to enrol 100 participants over the next six months and will publish the results openly. If you are a clinician, researcher, or public health worker interested in collaborating on this study, or if you are a community health organisation that would like to pilot the HealthAI application in your setting, we would very much like to hear from you. Two and a half million Indians develop TB every year. Fifty-seven percent of Indian diabetics don't know they have it. The diagnostic tools to close those gaps may already be in your hand.   About the author: HealthAI Labs Research team , an AI health screening platform based in India.  The HealthAI application is available at www.gethealthai.in.