IIT Madras and CMC Vellore Develop AI Tools to Detect and Assess Kidney Disease

Researchers at the Indian Institute of Technology Madras have partnered with doctors from Christian Medical College, Vellore, to develop artificial intelligence tools designed to assess kidney disease risk, interpret CT scans, and measure kidney tumours.

Chronic kidney disease, which impairs the organ's ability to filter waste and excess fluid from the blood, often develops unnoticed in its initial stages. Patients are frequently diagnosed only after significant damage has occurred, leaving advanced cases reliant on dialysis or organ transplants. The new tools are aimed at early detection and assessment, addressing conditions highlighted by recent research including a 2025 ICMR-INDIAB study of 25,408 people that found impaired kidney function in 3.2 percent of participants, as well as a 2025 study of 3,350 Tamil Nadu agricultural workers that identified chronic kidney disease in 5.31 percent of subjects.

The collaborative team created three distinct tools targeting different clinical stages. The first evaluates clinical and laboratory information to calculate an individual's chronic kidney disease risk. After testing four computational techniques, the team selected a random forest model, which synthesises multiple decision paths to spot disease patterns. This model was developed using a public dataset of 400 patient records—comprising 250 individuals with the disease and 150 without—and evaluated 26 clinical and laboratory variables. The researchers also created a prototype interface to assist medical practitioners in using the system.

The second tool reviews kidney CT scans, using automated processing trained on approximately 12,400 publicly available kidney images. The system is designed to classify scans into four categories: normal, cyst, stone, or tumour.

The third tool generates three-dimensional reconstructions of the kidney and tumour directly from CT scans to calculate tumour volume and disease burden. In the cases analysed, kidney volumes ranged between 120 mL and 245 mL, tumour volumes ranged from 2 mL to 24 mL, and tumour burdens ranged from 1 percent to 10.6 percent.

Professor G L Samuel of the Department of Mechanical Engineering at IIT Madras explained that the ultimate goal is to create a digital twin of the kidney trained on patient data to simulate progression. If scan and clinical data are fed into the system, Samuel noted, it could help clinicians understand how the condition might progress over three, six, or twelve months if left untreated.

Dr Santosh Varughese, a nephrologist at CMC Vellore, noted that the tools draw upon data commonly available in community, primary-care, and general-practice environments to assist in prioritising patients who require nephrology referrals or comprehensive evaluations. The technologies currently remain at the research stage. The researchers plan to gather additional patient data and validate the models against clinical cases over the next two years, with ethical clearances and hospital deployment expected to take approximately five years.