Ethical AI in Biomedical Informatics

Abstract

Biomedical AI systems increasingly influence clinical decisions -- from diagnostic imaging interpretation to treatment recommendation to risk stratification -- yet ethical frameworks for governing these systems remain fragmented across regulatory bodies, professional societies, and technology developers. The stakes are uniquely high: a biased diagnostic algorithm can systematically misdiagnose an entire demographic group, an opaque risk model can deny care without explanation, and a privacy breach in genomic data can expose an individual's disease predisposition to insurers and employers. We present the Ethical AI in Biomedical Informatics Framework (EABIF), evaluating five ethical governance approaches -- algorithmic fairness auditing, explainable AI for clinical transparency, privacy-preserving computation, participatory design with patient communities, and regulatory compliance frameworks -- across four biomedical AI domains (clinical decision support, genomic medicine, medical imaging AI, and digital health monitoring). Our Ethical AI Governance Score (EAGS) measures fairness achievement, transparency quality, privacy protection strength, stakeholder inclusion, and regulatory alignment. Participatory design achieves the highest EAGS (0.926) by centering patient and community voices in AI development, while algorithmic fairness auditing achieves the highest measurable equity impact (0.960) through systematic bias detection and mitigation across demographic groups.