New Delhi: Speaking at the sixth edition of the ET Healthcare Leaders Summit, experts said India has no shortage of AI pilots in healthcare, however, the real challenge is getting them past pilot stage, and getting hospitals to see them as a cost saver rather than a fresh expense.
The discussion was part of a panel titled “Is AI the Care Catalyst Expanding Reach, Scaling Impact? ” featuring J.P. Dwivedi, Chief Information Officer (CIO), Rajiv Gandhi Cancer Institute and Research Centre, Dr. Saurav Basu, Senior Scientist, Indian Council of Medical Research (ICMR), and Kalyan Sivasailam, Co-Founder, 5C Network.
The pilot problem
Dr Basu said India already has plenty of AI pilots across screening, diagnostics and surveillance, many of them operational in states that have moved fast on digital health. But he flagged a persistent gap between screening and actual care.
“A screening might provide you with a diagnosis, but algorithms don’t provide care, healthcare systems do,” he said. “Are the healthcare institutions sufficiently equipped to provide the necessary care once the early screening and likely diagnosis has been made? That would be the fundamental challenge.”
He pointed to what he describes a “valley of death,” where an estimated 95 per cent of pilots never get integrated into the health system, because they are developed in silos without accounting for how they will function in low-resource settings.
Cautioning against treating scale and equity as the same goal, Dr Basu said “Algorithmic biases usually creep in when the datasets used for training are not particularly representative of the population. India’s diversity – socioeconomic, ethnic, cultural and linguistic; makes this a major challenge. Often, pilots show us an accuracy of 90 per cent, but that’s on a very small sample, in just one setting. Those are red flags for me.”
Distinction between software and AI
For Sivasailam, the distinction between AI and conventional software is what determines whether it can scale commercially.
“AI has to be deflationary, not inflationary in nature. Software is something that adds cost. You need to buy it, train people, then hire more people to work on it. AI is almost a form of labour, and it can potentially be much more deflationary,” he said.
He described AI as functioning like “an infinite resident” for radiologists which can reduce the number of decisions and clicks required per case rather than adding to them.
Informing that this shift is being adopted faster by smaller institutions than by large ones, Sivasaialam, said, “The smaller hospitals and diagnostic centres are much more open and really require AI. It’s a little bit like how the US is still in a kind of check economy, while India has moved on to a total digital, 5G kind of ecosystem.”
He added that many of these smaller centres proactively reach out for newer models the moment a regulatory certification comes through.
On where the technology is headed next, Sivasailam said AI in radiology is moving beyond its early focus on screening.
“There is this trend around talking about AI in radiology from a screening perspective; TB screening, lung cancer screening. But the capabilities of newer architectures today are allowing us to be much more ambitious about what AI can do, not in screening, but in diagnosis.I think India will probably be the best in the world at it, because of our scale of data, which allows us to build much more generalisable models.”
Clinical adoption no longer a hurdle
Dwivedi said clinical adoption is no longer necessarily AI’s biggest barrier.
“For a change, this is the first time that clinical adoption is pulling us rather than pushing back,” he said.
He highlighted practical applications ranging from automated classification of patient records to AI-generated clinical summaries and radiology trend analysis. At his institution, AI-assisted document processing has reduced a task that previously took several minutes to about a minute, he said.
However, Dwivedi cautioned against fragmented adoption echoing Dr. Basu’s point on integration.
“Approaching AI from a siloed perspective is very dangerous. It has to be one thought-through, well-orchestrated AI platform.”
Panelists agreed that the real test of AI is whether the technology can actually reach more patients with better diagnosis and specialist care, without cutting corners on safety, cost, or how well it fits into the way hospitals actually work.
