14496
clinical study

Validation of a Deep Learning Algorithm in Detecting Cervical Spine Fractures on CT

Materials & Methods
This multi-institutional retrospective study evaluated the diagnostic performance of a commercial deep learning algorithm for detecting acute cervical spine fractures on CT. A total of 2,653 adult CT cervical spine examinations performed between Nov. 2020 and March 2021 were analyzed. AI outputs were compared against original radiologist interpretations. All discrepant cases and all AI-positive cases underwent adjudication by a panel of three radiologists to establish ground truth. Sensitivity, specificity and concordance rates between AI and radiologist interpretations were calculated.

Results
The prevalence of acute cervical spine fracture was 4.1%. Radiologists demonstrated a sensitivity of 95.4% and specificity of 99.7%, while the AI algorithm achieved a sensitivity of 88.9% and specificity of 99.1%. Overall concordance between AI and radiologist reports was 93.5%. Of 108 confirmed fractures, AI correctly identified 96 cases, with 22 false positives. Common sources of false positives included chronic fractures and beam-hardening artifacts.

Conclusions
The AI algorithm demonstrated high specificity, with fracture detection performance comparable to radiologists in specificity but inferior in sensitivity. These findings support AI’s role as a complementary triage and safety tool rather than a replacement for radiologist interpretation in cervical spine trauma imaging.

Disclaimer: This links to a third-party website, which may require credentials to review.

Explore the Latest AI Insights, Trends and Research