AI can lift care where specialists are scarce, says Soumya Swaminathan
New Delhi: Former WHO Deputy Director-General Soumya Swaminathan said AI can materially improve healthcare access in places short on specialists, citing India and parts of Africa. Image and pattern recognition-reading X-rays and pathology slides-are the clearest near-term wins, provided models are trained on high-quality, representative datasets.
These tools are already in circulation, she noted, with many more on the way. Her caution to health systems: treat new AI like a new drug or vaccine-prove efficacy and safety before scale, and bring it under a clear regulatory pathway.
Where AI helps right now
- Radiology support: triage chest X-rays, flag likely TB or pneumonia, prioritize reads during surges.
- Pathology pre-screening: detect suspicious regions on slides, reduce false negatives, speed up workflows.
- Mental health: assist screening and follow-up using validated tools, with clinician oversight.
- Frontline augmentation: decision support in low-resource settings to standardize basics and escalate the right cases.
Clinical standard, not hype
Swaminathan urged formal evaluation before deployment at scale-like trials for drugs and vaccines. That means real-world performance data, safety monitoring, and a defined path to authorization and post-market surveillance.
- Run prospective, clinically relevant studies with blinded comparison to gold-standard reads.
- Report sensitivity/specificity, AUC, NPV/PPV by subgroup (age, sex, comorbidities, device type, site).
- Verify dataset quality and representativeness; document labeling standards and inter-rater agreement.
- Keep a human in the loop, with clear override rules and audit trails.
- Plan for data protection, consent, bias audits, and periodic re-validation after updates.
- Map to a regulatory route and post-market monitoring plan. See WHO's guidance on regulatory considerations for AI in health.
Implementation checklist for hospitals and programs
- Define a narrow, high-value use case with measurable outcomes (e.g., time-to-report cut by 30%, missed findings ↓).
- Assess infrastructure: PACS/LIS integration, device compatibility, data security, offline/edge needs.
- Procure transparently: demand model cards, validation evidence on local populations, and support SLAs.
- Pilot in one unit, measure drift and equity impacts, and compare against current standard of care.
- Train clinicians and technicians; standardize escalation and documentation inside existing workflows.
- Monitor continuously: false positives/negatives, turnaround times, clinician trust, patient outcomes, and costs.
India AI Impact Summit 2026: headlines clinicians should note
- Guiding pillars: People, Planet, Progress-human-centric benefits, environmental responsibility, and inclusive growth.
- MANAV Vision announced: Moral and Ethical Systems, Accountable Governance, National Sovereignty, Accessible and Inclusive, Valid and Legitimate.
- Infrastructure: Tata Group and OpenAI to build 100 MW of AI capacity in India, scalable to 1 GW.
- Models: BharatGen Param2 (17B parameters; 22 languages) launched, alongside new LLMs from Sarvam AI.
- Public interest was strong; the India AI Impact Expo was extended and closed on February 21, according to ANI.
Why this matters for healthcare: stronger compute and multilingual models can push validated tools closer to the point of care-especially where specialists are scarce. But adoption must follow clinical evidence, tight integration with PACS/LIS, and ongoing safety checks.
For policy and governance context, see WHO's ethics and governance of AI for health. For practical training and implementation insights across diagnostics and care pathways, explore AI for Healthcare.
Your membership also unlocks:
