Human Oversight Key to Safe Use of AI in Healthcare, Pg20

Key Highlights:

• Researchers at IIT-Madras developed Garbhini-GA2, an AI model that predicts foetal age using ultrasound scans, with a mean error of only half a day.
• Hadlock’s formula, the current global standard, can misjudge foetal age in Indian contexts by up to 7 days.
• ARMMAN NGO deployed an AI chatbot to support Auxiliary Nurse Midwives (ANMs) in managing high-risk pregnancies; received 94% positive feedback.
• Virtual autopsies using CT/MRI scans reduce the invasiveness and time of postmortem processes and improve public acceptance.
• AI tools can diagnose causes of death like drowning or cerebral haemorrhage with accuracy up to 92%.
• Data privacy and weak regulation under current laws (IT Act 2000 and DPDP Act 2023) are major concerns.
• Automation bias and speech recognition issues in regional languages challenge effective AI deployment.

Detailed Insights:

• AI applications like Garbhini-GA2 address racial bias in global medical tools by using population-specific datasets from India.
• High-risk pregnancies (HRPs) (nearly 50% of Indian pregnancies) require regular monitoring, where AI chatbots help overcome healthcare personnel shortages.
• ANMs in rural areas benefit from AI assistance, though limitations in voice recognition for diverse Indian languages remain a barrier.
• Virtual autopsies (virtopsies) reduce delays in funeral rites and address cultural sensitivities associated with body mutilation in conventional autopsies.
• CNN-based AI models improve forensic outcomes by classifying causes of death through image analysis with high precision.
• Automation bias undermines expert judgment; even experienced radiologists were influenced by incorrect AI-generated scores in controlled studies.
• India’s legal framework lacks clarity on AI accountability and data processing, necessitating regulatory reforms and stronger governance mechanisms.
• Robust data governance, clinician training, and clearly defined roles for AI are crucial for safe, ethical integration in healthcare delivery.

Scientific/Technical Concepts Involved:

• Garbhini-GA2: AI-based model for estimating foetal age from ultrasound scans using Indian population data.
• Hadlock’s Formula: Standard foetal dating method based on Western datasets.
• Convolutional Neural Networks (CNNs): Deep learning models used for image-based classification in medical imaging.
• Virtual Autopsy (Virtopsy): Non-invasive postmortem analysis using imaging techniques like CT and MRI.
• Automation Bias: Cognitive bias where humans over-rely on automated systems, even when they are inaccurate.

Mains Mock Question:
Discuss the benefits and risks associated with integrating Artificial Intelligence in healthcare delivery in India. How can human oversight be institutionalised to ensure ethical and effective use of AI?