An international team of radiology experts examined the critical need for robust monitoring systems to ensure AI safety, efficacy, and regulatory compliance in real-world radiology practice. Published in the Korean Journal of Radiology, the perspective paper addresses the gap between regulatory frameworks and practical implementation of AI monitoring systems. The authors analyzed current regulatory frameworks including FDA guidance, European AI Act requirements, and international standards from organizations like the American College of Radiology and the European Society of Radiology.
The paper outlines comprehensive monitoring strategies across three key domains: technical performance (tracking algorithm accuracy, detecting data drift, and monitoring system failures), network infrastructure (ensuring complete data transfer from modalities to AI systems), and human-AI interaction (capturing radiologist feedback and mitigating automation bias). Real-world examples demonstrate the value of monitoring dashboards that track adoption rates, AI-radiologist agreement, technical errors, and turnaround times; enabling data-driven decisions such as discontinuing poorly performing tools or adjusting workflow orchestration. The authors emphasize that passive monitoring alone is insufficient, advocating for real-time alerting systems with performance thresholds that flag algorithmic deviations and enable rapid intervention to maintain patient safety and diagnostic quality.