AI in Cervical Spine Fractures-Initial Experience

Purpose: Cervical spine fractures can occur secondary to exaggerated flexion or extension, or because of direct trauma or axial loading. Our aim was to evaluate the accuracy of a commercially available AI algorithm in suspected cervical spine fractures.

Methods: During one month period 852 CT cervical spine exams where analyzed by the AI algorithm, out of that 38 cases where flagged as fractures (3.99%). 2. A senior neuroradiologist evaluated all cases in order to access how many where true positive (TP) acute fractures vs false positive (FP). TP fractures were defined as acute fractures, while FP were defined as chronic fractures or any other reason that the AI algorithm flagged the exam.
Results:

All cases submitted for analysis underwent evaluation by the AI software within 5-20 minutes. Out of the 38 cases 14 were TP acute fractures (37%), while 24 cases (63%) where FN , such as 1. chronic fractures 2. lucent lines traversing the bone, i.e, a nearby small vessel 3. artifact from shoulders 4. sesamoid osteophyte, 5. dural calcifications 6. post -surgical changes 7. partially calcified nearby vessels as well as 8. calcific tendinitis of the longus colli.

Conclusion: The AI algorithm flagged true fractures as well as other pathologies/variants or artifacts that were not acute fractures. This highlights the true complexity of cervical spine fractures. There is still need to evaluate the cases that were not prioritized to evaluate the existence of non-detected fractures.