AI-based model Garbhini-GA2

Researchers from IIT-Madras and THSTI Faridabad have developed an AI-based model called Garbhini-GA2 to improve foetal age estimation using ultrasound scans. Trained on data from Indian women, it significantly outperforms the conventional Hadlock’s formula, which is based on Caucasian data and often misestimates foetal age in Indian populations.

Garbhini-GA2

• Developed collaboratively by IIT-Madras and THSTI.
• Uses machine learning to estimate foetal age from ultrasonography data.
• Trained on labelled data from ~3,500 pregnant women in India.

Scientific Foundation and Data Source

• Model trained using annotated ultrasound scans: size, weight, and body parts of the foetus.
• Testing conducted on:
  • 1,500 scans from Gurugram Civil Hospital (not included in training).
  • 1,000 scans from Christian Medical College, Vellore.

Comparative Accuracy with Global Benchmarks

• Garbhini-GA2 average error: ~0.5 days.
• Hadlock’s formula error margin in India: up to 7 days.
• Hadlock’s inaccuracy stems from its reliance on data from Western populations, not accounting for Indian anthropometric variations.

Significance of AI in Healthcare:

• AI algorithms (especially deep learning models) are now able to interpret medical imaging such as X-rays, CT scans, and MRIs with accuracy comparable to expert radiologists.
• Example: A 2020 study in Nature showed AI outperformed radiologists in reducing false positives (by 1.2%) and false negatives (by 2.7%) in breast cancer diagnosis.
• Fields benefiting: Radiology, pathology, ophthalmology, dermatology.

• AI integrates data from genomics, lifestyle, and medical history to create customized treatment protocols.
• Example: IBM Watson Oncology has helped oncologists in 230+ hospitals globally by offering evidence-based treatment options.
• Reduces adverse drug reactions and enhances patient-specific therapy effectiveness.

• AI shortens drug development cycles by predicting drug-target interactions and molecule structures.
• Example: Insilico Medicine developed a fibrosis drug candidate using AI in just 46 days, bypassing years of traditional research.
• Potential to lower R&D costs and expand access to rare disease drugs.

• Natural Language Processing (NLP) tools can transcribe and summarize consultations, easing the burden on clinicians.
• AI optimizes appointment scheduling, resource utilization, and patient flow.
• Leads to reduced wait times, better hospital management, and improved work-life balance for healthcare workers.

• AI-enabled IoT and wearable devices track vital signs, enabling early intervention.
• Platforms like Babylon Health and WHO’s Sarah use AI chatbots for triaging and health promotion.
• Crucial for rural and underserved areas with limited access to specialists.

• AI-integrated VR/AR systems allow immersive surgical training with haptic feedback.
• Companies like FundamentalVR simulate real surgeries, enhancing precision and skill retention.
• Adaptive learning tools personalize content for faster and deeper learning in medical education.

Challenges of AI in Indian Healthcare

• Only 45% of India’s Health & Wellness Centres in rural areas have electricity backup, hindering AI deployment.
• Absence of high-speed internet and digital medical equipment in rural India is a major hurdle.

• Lack of standardized, high-quality Electronic Health Records (EHRs) and poor interoperability across systems.
• No national guideline for data retention or centralized health data integration.
• Inconsistent data hinders AI training and may cause model inaccuracies.

• As of 2023, 45% of India’s population lacked internet access (IAMAI-Kantar report).
• Urban-rural digital inequality may lead to AI benefits being concentrated in cities, widening the healthcare gap.

• The proposed DISHA Act (2017) to govern digital health data remains unpassed.
• Lack of regulations on AI validation, liability, patient rights, and data protection undermines trust and safety.
• Absence of legal clarity discourages large-scale public-private AI collaboration.

• Use of foreign-trained AI models raises applicability concerns in Indian socio-cultural contexts.
• Issues include algorithmic bias, informed consent, health data privacy, and misdiagnosis risk in multilingual, multicultural settings.

• AI system deployment costs range from USD 20,000 to USD 1 million.
• India’s public health expenditure remains low at 8% of GDP (2020-21).
• Smaller hospitals and clinics may find it financially unfeasible to adopt AI technologies.

• India has 22 official languages and 100+ dialects, making it difficult to build universally effective AI interfaces.
• AI chatbots or systems trained in one language may misinterpret or fail to serve patients in vernacular languages, risking safety and usability.