Validation and impact of AI tools for fracture detection in radiology

Abstract

Bone fractures represent one of the most common and rising global health burdens, with delays or errors in diagnosis leading to significant patient harm and healthcare costs. Radiologists face increasing workloads, making timely and accurate fracture detection ever more challenging. Artificial intelligence (AI) has emerged as a promising solution, with several tools developed to assist in musculoskeletal imaging. Yet, many existing studies rely on enriched datasets and limited validation, raising questions about their real-world applicability across diverse patient populations and healthcare systems.<br/>This thesis evaluates AI tools for fracture detection through a series of validation and impact studies. First, it reviews the most promising AI applications in musculoskeletal imaging and their potential clinical translation. Second, it presents external validation of AI models trained on data from different regions, revealing robust overall diagnostic performance but variable accuracy across anatomical subgroups and hospitals. Third, it assesses the impact of AI on clinical decision-making in multi-reader, multi-case studies, showing that AI improves diagnostic certainty and benefits less experienced clinicians most.<br/>The findings highlight AI’s potential to improve diagnostic accuracy, confidence, and workflow efficiency in fracture detection while underscoring the importance of local validation, robust reference standards, and prospective trials. By addressing key gaps in generalizability and real-world impact, this work contributes to a better understanding of how AI can be responsibly integrated into radiology practice to support clinicians and enhance patient care.<br/>

Ähnliche Arbeiten