International Multicenter Validation of an Expanded AI Diagnostic System for 18 Pathologies in Thoracic and...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Conventional radiography has high error rates (3-10%) due to clinical workload.
• An evidence gap exists regarding integrated AI systems for skeletal and thoracic abnormalities.
• AI shows promise in supporting radiographic diagnoses.

Purpose of the Study:

• To validate a unified radiographic AI suite for expanded diagnostic scope.
• To confirm the AI system's continued robustness in detecting pathologies.
• To assess AI performance across diverse patient demographics and international settings.

Main Methods:

• Retrospective evaluation of 21,581 adult and pediatric X-rays from 20 countries.
• Reference standard established by expert radiologists with adjudication.
• Calculation of diagnostic metrics (AUC, sensitivity, specificity) for 18 pathologies.

Main Results:

• For expanded scope findings, AUC exceeded 96.1%; sensitivity 94.5-98.8%; specificity 86.6-96.1%.
• For historically validated findings, AUC remained >96.1%; sensitivity 94.5-97.8%; specificity 84.6-89.4%.
• Consistent performance observed across patient age, sex, and country subgroups.

Conclusions:

• Deep learning can transition from narrow tools to unified, high-performance diagnostic systems.
• The validated AI suite demonstrates potential for broad clinical application.
• AI integration can enhance diagnostic accuracy and efficiency in radiography.