Machine Learning to Detect Cervical Spine Fractures Missed by Radiologists on CT: Analysis Using Seven Award-Winning...

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Previous studies on missed cervical spine fractures relied on human reevaluation, limiting the scope of error detection.
• Machine learning (ML) offers a novel approach to systematically identify fractures overlooked in initial interpretations.

Purpose of the Study:

• To leverage ML models for detecting radiologist-missed cervical spine fractures on CT scans.
• To characterize the nature and assess the clinical significance of these ML-identified missed fractures.

Main Methods:

• Retrospective analysis of adult emergency department cervical spine CT exams (2018-2022).
• Seven ML models identified potential missed fractures in initially negative reports.
• Neuroradiologists and spine surgeons reviewed ML-detected fractures for accuracy and clinical significance.

Main Results:

• Of 6378 initially negative exams, 356 showed ML-detected fractures; 40 were confirmed as true missed fractures in 39 patients.
• Commonly missed fracture sites included C7 transverse process and C5/C6 spinous processes.
• 15 of 40 missed fractures were deemed clinically significant, requiring further management.

Conclusions:

• ML models effectively identify cervical spine fractures missed by radiologists.
• Characterization of these misses can inform targeted quality improvement initiatives and enhance radiologist training.