IITs unveil low-cost AI tools to detect cancer and kidney disease via simple breath tests and oral swabs

Called MammoX, the platform provides a confidence score alongside visual cues, allowing radiologists to make informed decisions rather than relying solely on generic templates. “We draw data from hospital PACS (Picture Archiving and Communication System). MammoX differs from other AI systems, such as those used in CT scan analysis at AIIMS, because it is specifically tailored for mammography. Its pre-processing pipeline ensures high-quality input data, correcting images from different machines to maintain homogeneity for accurate AI analysis,” says Yadav.

Yadav emphasises that the AI identifies masses and calcifications with detailed descriptions. “The project, Radiologist AI, is being conducted in collaboration with Max Hospital Saket under the supervision of Dr Bharat, head radiologist, and Dr Shruthi Jadhav, professor at IIT Bombay. All tools developed are validated through this partnership,” says Yadav. He has also developed an orthopaedic surgery planner for knee replacement procedures. Traditionally, surgeons manually identify measurements and angles, a process that can take 25–30 minutes. Med Access automates this analysis using AI in just 60 seconds.

AI-powered microscope for early oral cancer detection

What if a simple oral swab could detect oral cancer, one of the deadliest of cancers, early? Sher Mohammad, a junior research fellow from the Indian Institute of Engineering Science and Technology, Shibpur (IIEST), has created a simple, affordable and non-invasive diagnostic tool to do just that. Working in collaboration with the Indian Institute of Technology (IIT), Kharagpur, Sher’s device can be deployed in rural and resource-limited healthcare settings.

It uses a multispectral microscope integrated with machine learning. A basic oral swab is collected from the patient’s mouth, and a smear is prepared on a glass slide. This is then analyzed under a multispectral microscope, which passes different wavelengths of light through the cells. They light up, generating distinct cellular images that can be captured using a camera or even a mobile phone for subsequent analysis. The system categorises these images into cancerous, smoker-related (potentially pre-malignant), or non-smoker/normal groups. “A healthcare worker or laboratory technician can prepare the smear, run the test and monitor results directly on a mobile phone interface,” he says.

The team collected around 50 patient samples from government medical colleges and hospitals across West Bengal, including Kolkata, achieving an overall accuracy of approximately 88 per cent. The tool is advancing into clinical trials.

AI-powered breath analyzer

The compounds in the air you breathe out can now act as biomarkers of multiple diseases, including chronic kidney disease, chronic liver disease and diabetes. Dr Tarun Kanti Bhattacharya, Institute Chair Professor and Head of the Department of Electronics and Electrical Communication Engineering at IIT Kharagpur, and his team have developed an AI-enabled breath analyzer aimed at transforming early disease diagnosis.

A hand-held, affordable diagnostic tool, a user simply breathes into its sensor, which generates time-series signals. These are then processed by an AI platform that extracts concentrations of various gases present in the breath. While a normal breath consists mostly of nitrogen and carbon dioxide, exhaled air also contains minute amounts of VOCs such as ammonia, acetone and formaldehyde. “The World Health Organization (WHO) has linked specific combinations of such trace VOCs to certain diseases — for instance, elevated ammonia levels may indicate kidney dysfunction, while acetone can signal diabetes,” says Dr Bhattacharya.

“A synthetic dataset forms the basis for initial training, after which actual patient breath signals, collected ethically from hospital outpatient departments, are used to fine-tune and calibrate the neural models,” he adds. According to Dr Bhattacharya, this two-stage approach helps achieve reliable AI decision-making even in the presence of comorbidities — where multiple diseases co-exist and VOC patterns become more complex.

“Unlike invasive diagnostic procedures, the breath analyser delivers a comprehensive metabolic profile within seconds of a breath sample. The sensor, comparable to a strip in a glucometer, is replaceable after multiple uses,” he says. Clinical trials are under way at institutions like Calcutta Medical College and North Bengal Medical College. As the device does not enter the body, its regulatory pathway is similar to that of thermometers and blood pressure gauges.

Future vision

Keratoconus is a rare but progressive eye disease in which the cornea, the clear, dome-shaped front surface of the eye, gradually thins and bulges outward into a cone shape. This distortion prevents light from focussing properly on the retina, leading to blurred vision and, in severe cases, significant sight impairment.

“It affects roughly three in one lakh people. Because it is rare, it is often the last condition clinicians suspect. By the time other conditions like glaucoma or cataract are ruled out, precious time has already been lost, especially in children,” says Dr Nagarajan Ganapathy from IIT Hyderabad’s biomedical engineering department. His device, Keratofy, uses an infrared light spectrum combined with a near-infrared camera. Because the light operates in the infrared range, it does not cause discomfort or visible glare for the patient. The reflected light patterns are captured and analysed instantly. “In a normal eye, when light enters the cornea, it distributes evenly. In a keratoconus-affected eye, the cornea becomes conical, so the light distribution becomes uneven. That is the key difference we capture,” he says. “It is an edge-based system, meaning the analysis happens directly on the device, with no internet connectivity required,” he adds.

The team collaborated with LV Prasad Eye Institute in Hyderabad for data collection and clinical validation. “We screened around 400 patients and used that data to train and refine the algorithm,” he explains. “Our current validation shows approximately 96 percent accuracy in detecting keratoconus cases.” The simple camera module costs around Rs 4,000.