IIT Kanpur’s BODH AI could make hospital data gathering safer, surer: Here’s how it protects patients

“Currently, many AI models show high accuracy during development but these results are often based on limited or carefully curated datasets. When deployed in hospitals, where patient populations are more diverse and conditions more variable, performance can differ significantly. This gap between controlled testing and real-world use has been identified as a key challenge in the adoption of healthcare AI,” explained Professor Nisheeth Srivastava, who conceived the project with a team of five, who worked on the project for 18 months.

“The biggest challenge with health AI today is not innovation, but trust,” said Professor Srivastava. Bodh allows access to reliable test data.

Safe and trustworthy data

Traditionally, independent auditors need access to large datasets to verify whether an AI system is safe and effective. In healthcare, however, such data is difficult to obtain due to privacy concerns.

BODH AI offers a different approach. Instead of moving sensitive patient data, the system sends AI models to secure data environments. The models are tested where the data already exists, and only the performance results are shared. This allows evaluation without exposing personal health information.

At the same time, developers using the platform contribute to building a pool of test data that can later be used by the National Health Authority for independent assessments.

Consent-based data sharing
“The framework also builds on the consent-based data-sharing system under the Ayushman Bharat Digital Mission. Patients can permit the use of anonymised health data for research and testing purposes, extending the use of digital health records beyond clinical care while retaining individual control over data access,” Professor Srivastava said.

Quality and reliability

Another component of the system focuses on maintaining the quality and reliability of datasets used for testing. Repeated use of the same datasets can reduce their effectiveness in evaluating AI systems. BODH AI incorporates mechanisms to ensure that testing data remains statistically valid over time, with the aim of producing more consistent and realistic performance assessments.

Why it matters

The system is particularly relevant in India, where healthcare data is often fragmented across different hospitals and regions. By enabling secure access to diverse datasets, BODH AI aims to improve how well AI systems work across varied populations.

For patients, the benefits may not be immediately visible but could be significant. AI tools are increasingly used to assist in diagnosis and treatment decisions. Poorly tested systems can lead to errors, while well-evaluated tools can improve clinical outcomes.

BODH AI also opens the door to greater transparency. Over time, it could allow comparisons between different AI tools and support regulatory oversight through standardised benchmarks.

“We expect BODH AI to begin deployment in government hospitals by the end of this year,” Professor Srivastava said.