Enterprise AI Assurance Framework for Biomedical Diagnostics: A Human-Centered Approach for Trustworthy, Explainable, and Safe Artificial Intelligence Systems

<jats:p>Artificial intelligence (AI) has demonstrated significant potential for transforming biomedical diagnostics through advanced pattern recognition, predictive modeling, and clinical decision support. However, the adoption of AI systems in healthcare remains challenged by limitations associated with explainability, reliability, safety, fairness, human confidence, and continuous operational monitoring. Current AI evaluation approaches primarily focus on predictive performance metrics, such as accuracy and sensitivity, while providing limited mechanisms for assessing whether AI systems are trustworthy, interpretable, and suitable for real-world clinical environments. This study proposes the Enterprise AI Assurance Framework for Biomedical Diagnostics (EAIA-BD), a human-centered framework designed to support the development, evaluation, deployment, and continuous governance of trustworthy AI-enabled biomedical diagnostic systems. The proposed framework introduces an AI Assurance Intelligence Layer that extends beyond conventional model evaluation by integrating reliability assessment, explainability analysis, safety evaluation, fairness assessment, human confidence considerations, and continuous monitoring capabilities. The framework introduces two key contributions: (1) the Human Validation Boundary, which establishes a structured interaction point between AI-generated recommendations and clinical decision-making, ensuring that healthcare professionals retain validation authority and accountability; and (2) the Biomedical AI Assurance Score (BAAS), a multidimensional evaluation model that measures AI system readiness across reliability, explainability, safety, fairness, human confidence, and monitoring dimensions. The proposed approach is developed using a design science methodology and evaluated through expert-oriented assessment, scenario-based analysis, and comparison with traditional AI evaluation approaches. The framework provides a systematic pathway for transitioning biomedical AI systems from accuracy-focused models toward trustworthy, explainable, and continuously assured clinical intelligence systems. This research contributes a reference model for organizations seeking to responsibly adopt AI in biomedical diagnostics by integrating technical performance, human-centered validation, and lifecycle-based assurance into a unified framework.</jats:p>