همایش بین المللی هوش مصنوعی

صفحه اصلی / دومین همایش بین المللی هوش مصنوعی

From Code to Clinic: Bridging the Implementation Gap in AI-Based Clinical Decision Support

نویسندگان :

Ehsan Ghazanfari Savadkoohi¹ Farkhondeh Asadi² Hassan Emami³

1- دانشگاه علوم پزشکی شهید بهشتی 2- دانشگاه علوم پزشکی شهید بهشتی 3- دانشگاه علوم پزشکی شهید بهشتی

کلمات کلیدی :

Artificial Intelligence،Clinical Decision Support Systems،Implementation Science،Algorithmic Stewardship،MLOps،Translational Gap،Socio-technical Systems

چکیده :

Background: Artificial intelligence-based clinical decision support systems (AI-CDSS) offer a method to augment human cognition with data-driven foresight. A contradiction defines the field. While high-performing algorithms appear exponentially in the literature, real-world adoption remains negligible. Objectives: This narrative review examines the "implementation gap." We identify why technically sound models fail in practice and propose a framework for sustainable translation. Methods: We synthesized evidence from empirical deployment studies, systematic reviews, and implementation science frameworks (CFIR, RE-AIM) published between 2015 and 2025. The analysis maps barriers across technical, human, and systemic domains. Results: Failure to scale is rarely caused by algorithmic insufficiency. It stems from socio-technical friction. Key barriers include fragmented data infrastructure, a lack of prospective cross-site validation, and workflow misalignment that triggers alert fatigue. Systemic hurdles, including liability ambiguity and incompatible reimbursement models, further disincentivize institutional adoption. We present an integrated roadmap. It prioritizes "system-centric" validation over "model-centric" optimization. Strategies include hybrid effectiveness-implementation trials, "silent trial" calibration, and algorithmic stewardship for continuous surveillance (MLOps). Conclusions: Sustainable AI-CDSS implementation requires a shift. We must stop treating AI as a static product and start managing it as a complex adaptive service. Success depends on rigorous governance, human-centered design, and incentive alignment.