Part 2: AI in Healthcare | Module 11

🏥 Healthcare AI Landscape

AI in healthcare encompasses a wide range of applications from diagnostic imaging to drug discovery, each with distinct regulatory pathways and risk considerations.

Key Application Areas

Diagnostic Imaging: AI analysis of X-rays, CT scans, MRIs, pathology slides
Clinical Decision Support (CDS): Treatment recommendations, risk assessment
Drug Discovery: AI-driven compound identification and optimization
Predictive Analytics: Patient deterioration, readmission risk
Administrative AI: Scheduling, billing, documentation automation
Remote Monitoring: Wearable device analytics, telemedicine AI

⚖ FDA Regulatory Framework

The US FDA regulates AI/ML-based Software as a Medical Device (SaMD) under a risk-based framework.

FDA Risk Classification

Class	Risk Level	Regulatory Pathway	Examples
Class I	Low Risk	General Controls (exempt)	General wellness apps
Class II	Moderate Risk	510(k) Premarket Notification	Most diagnostic AI
Class III	High Risk	PMA (Premarket Approval)	Novel high-risk devices

FDA AI/ML Software Guidance

Predetermined Change Control Plan (PCCP): Document planned modifications
Good Machine Learning Practice (GMLP): Quality principles for AI development
Real-World Performance: Post-market monitoring requirements
Transparency: Labeling requirements for AI devices

💡 FDA PCCP Framework

The Predetermined Change Control Plan allows manufacturers to specify anticipated modifications to AI/ML algorithms, enabling certain changes without requiring new premarket submissions if they fall within the approved plan.

🇮🇳 CDSCO (India) Requirements

India's Central Drugs Standard Control Organisation regulates medical devices including AI-based software.

CDSCO Classification

Class	Risk	Requirements
Class A	Low	Registration with state authority
Class B	Low-Moderate	CDSCO manufacturing license
Class C	Moderate-High	CDSCO license + clinical evidence
Class D	High	Full clinical investigation

⚠ Import Requirements

Foreign manufacturers must appoint an authorized Indian agent and register with CDSCO. All medical AI software used in India must comply with Medical Device Rules, 2017.

🇪🇺 EU MDR and AI Act

Medical AI in Europe must comply with both the Medical Device Regulation (MDR) and, for high-risk applications, the EU AI Act.

Dual Compliance Requirements

MDR 2017/745: CE marking, conformity assessment, post-market surveillance
EU AI Act: High-risk AI requirements apply to medical devices
GDPR: Health data processing requirements

CE Marking Process for AI Medical Devices

✓Classification under MDR Annex VIII
✓Quality Management System (ISO 13485)
✓Clinical evaluation and clinical investigation if required
✓Technical documentation per Annex II/III
✓Notified Body assessment for Class IIa and above
✓Post-market surveillance system

📈 Clinical Decision Support

Clinical Decision Support (CDS) systems provide healthcare professionals with knowledge and patient-specific information to enhance decision-making.

CDS Regulatory Considerations

CDS Type	FDA Regulation	Key Factors
Non-Device CDS	Not regulated as device	Practitioner can independently review basis
Device CDS	Regulated as SaMD	Time-critical, not independently reviewable

✅ CDS Five Criteria Test

FDA's non-device CDS criteria: (1) Not intended to acquire/analyze medical images/signals, (2) Intended for displaying/analyzing patient-specific information, (3) Provides recommendations, (4) Intended for healthcare professional review, (5) Basis for recommendations can be independently reviewed.

📚 Key Takeaways

1Healthcare AI is regulated as Software as Medical Device (SaMD) in most jurisdictions
2FDA uses risk-based classification with specific pathways for AI/ML devices
3CDSCO requires registration and licensing for medical AI software in India
4EU requires dual compliance with MDR and AI Act for high-risk medical AI
5Clinical Decision Support has specific criteria determining regulatory status