Machine Learning to Detect Cervical Spine Fractures Missed by Radiologists on CT

This study explored how machine learning (ML) models can be used to detect cervical spine fractures that were missed by interpreting radiologists on CT scans. The aim was to assess the nature and clinical significance of these missed fractures and provide insights that could help improve radiologists’ search patterns.

Researchers retrospectively analyzed 6979 CT scans from adult emergency department patients between 2018 and 2022. Seven award-winning ML models from the 2022 RSNA AI Challenge were used to flag scans with potential fractures not identified in original radiology reports. Scans marked positive by at least four models were reviewed by neuroradiologists, who used ML-generated heat maps to confirm true fractures. Two spine surgeons independently evaluated the clinical significance of confirmed misses.

ML identified 40 previously missed cervical spine fractures across 39 patients. These fractures often involved the C5, C6, and C7 levels, particularly the transverse and spinous processes. Surgeons deemed 37.5% of the missed fractures clinically significant, warranting MRI, surgical consult, CTA, or collar immobilization. This study underscores the potential of ML tools to catch subtle yet important fractures and their value in quality improvement and educational efforts within radiology.

This study was led by Drs. Amy Chen and Errol Colak (University of Toronto and Toronto Radiology), with co-authors Zixuan Hu, Kevin D. Shek, Jefferson Wilson, Fahad Saud S. Alotaibi, Christopher D. Witiw, Hui Ming Lin, Robyn L. Ball, Markand Patel, Shobhit Mathur, and Ervin Sejdić. Published in the American Journal of Roentgenology, Volume 224, Issue 3.

Link to Publication:
 https://www.ajronline.org/doi/10.2214/ajr.24.32076

Using Artificial Intelligence to Improve the Accuracy and Efficiency of Interpreting Thousands of Images in a CT Study – James S. Jelinek, MD​

How Machine Learning Detects Missed Cervical Spine Fractures and Can Be Used to Reduce Blind Spots – Tim S. Fischer, MD​