AI diagnosis: AI diagnosis is just what the doctor ordered

A man in his 70s came to the Jiwan Jyoti Christian hospital in Robertsganj, Uttar Pradesh, early this year, complaining of a fever that had not gone down in months, with no doctor offering an effective diagnosis. Dawn Kuruvilla, consultant physician at the hospital, too was initially befuddled. But a chest X-Ray he prescribed set off alarms on the qTrack dashboard, an AI application developed by Mumbai-based startup Qure.ai that uses deep learning to analyse chest X-rays for abnormalities. It revealed that the patient’s lungs were infected by TB bacteria, later confirmed by a CT scan.

In India, where the existing healthcare system faces enormous pressure, AI can be a big help as it can minimise oversight, make diagnostics more accurate, support clinician decision-making, answer patient queries and democratise healthcare at large.

As per WHO data, the density of doctors (number of doctors per 10,000 population) in India stood at 7.3 in 2020, compared to 17.2 worldwide, a decline from 12 in 1991. By contrast, China , had 25 doctors per 10,000 people as of 2021, up from 11 in 1990. The figures for developed countries such as the US, UK, Sweden and Australia stand at 36, 32, 52, and 40, respectively. Healthcare players and tech companies alike are investing in developing innovative AI solutions to address this gap.

While pharma firms like Pfizer and DRLare increasingly using AI for drug discovery or for its monitoring adverse effects, large hospital chains are leveraging AI tools to help predict the risk of various disease in patients. Apollo Radiology International for instance is using Google’s AI models to conduct 3 million free AI-powered screenings for TB and lung and breast cancer in underserved communities across India.

“AI is a tool and an enabler for healthcare solutions,” Bakul Patel, senior director, global digital health strategy and regulatory at Google told ET.

He, however, added that these solutions must be tailored to the local context, understanding local structures, rather than imitating what has been done elsewhere. For one, Google’s health acoustic representations foundation model trained on bioacoustic sounds can help researchers build models to detect particular diseases. Researchers at Microsoft partnered with Sankara Eye Hospital, to develop and deploy an ‘expert-in-the loop’ chatbot for cataract patients called CataractBot, available on WhatsApp in English, Hindi, Kannada, Tamil and Telugu. Dozens of startups are building AI -led tools in healthcare given the opportunity to make a positive difference. Startups like Qure.ai work with state governments like Karnataka and Maharashtra for early detection of lung cancer and tuberculosis. The company, founded in 2016, uses AI and deep learning to analyse chest X-rays that can detect lungs abnormalities. Srikanth Nadhamuni, CEO, Khosla Labs, and chairman of 10BedICU said it is piloting generative AI tool CARE that is voice-enabled and trained in the 10 major Indian languages. The firm has also developed tele-health ICU, where the shortage of specialists is handled by medical colleges over its cloud software. The company along with doctors from the government and private hospitals are coming up with standard ICU protocols for diseases that are common in their ICU like septic shock, he added. But there are downsides to scaling up such systems as well. Though India is known as a huge data generator, but data availability remains a challenge.

Non-profit Wadhwani AI is building AI solutions for TB detection, predicting patient follow-up likelihoods, and maternal and child care. Nakul Jain, director – product and solutions at Wadhwani AI, said that in secondary or tertiary clinics where data is collected, it may be lost as it is not stored in a planned and systematic way or is not usable for AI training.

“There needs to be an active effort to put these systems in place, to have a strategy around long-term data collection and storage,” Jain said. Another big challenge is last mile connectivity in a country like India. India is often not considered a huge market for many companies since cracking public health is complex for smaller startups.

In addition, India is a price sensitive market. Kuruvilla, who was cited earlier, shared that over reliance on tools could also be a concern, particularly for junior doctors. According to Kuruvilla, doctors should override AI diagnostics if they think it is necessary. While patient consent and data anonymisation are crucial, implementing regulation must be carefully and practically thought out as healthcare is a very large and unorganized industry and the space is still evolving, experts said.