BoneView Trauma

Assessment of an AI aid in detection of pediatric appendicular skeletal fractures by senior and junior radiologists.

Maarek et. al

ECR 2022

Published In

ECR 2022 (2022)

Authors

Richard Maarek A-L. Hermann, A. Kamoun, R. Khelifi, A. Marchi, M. Collin, A. Jaillard, T. Nguyen, H. Ducou Le Pointe;

Abstract

Purpose :
The number of conventional X-ray examinations in paediatric emergency departments is constantly increasing, which can lead to avoidable errors in interpretation by the radiologist. The use of artificial intelligence (AI) could improve the interpretation workflow by prioritising pathological radiographs as well as providing assistance in fracture detection.

Material and methods:
A cohort of 300 anonymized radiographs performed for peripheral skeletal fracture detection in patients aged 2 to 21 years was retrospectively collected. The gold standard was established for each examination after an independent review by two radiologists expert in musculoskeletal imaging. In case of disagreement, a consensual review with a third expert radiologist was performed. The 300 examinations included 60 exams per bodypart (hand/wrist, arm/elbow, shoulder, leg/knee, foot/ankle), half of which presented at least one fracture. All radiographs were then read by 3 senior pediatric radiologists and 5 radiology residents in training between the 2nd and 4th year of residency. The X-rays were read without the help of AI (BoneView, Gleamer) and immediately after with the AI help. Sensitivity and specificity for each group of radiologists were calculated without and with the help of AI.
The study was approved by an IRB n° CRM-2109-203.

Results:
The mean sensitivity for all groups was 73.3% without AI, it increased significantly by 9.5 points (95% CI: [7.05, 11.95], p<.001) to 82.8% with the aid of the AI. For the residents it increased from 71.9% to 82.2% (+10.3 points,  95% CI: [7.81, 12.72]; p< .001) and for the seniors radiologists from 75.6% to 83.8% (+8.2 points, 95% CI: [-2.42, 18.87], p=0.08). On average, the specificity increased from 89.6% to 90.3% (+0.7 points, 95% CI: [-0.75, 2.25], p=.28). For residents, it increased from 86.2% to 87.6% (+1.4 points, 95% CI: [-1.28, 3.95], p=.23) and slightly decreased from 95.1% to 94.9% (-0.2 points, 95% CI: [-1.18, 0.73], p=.42) for senior radiologists. The increase in sensitivity was similar between the children (2-12 years old) and the adolescent group (12-21 years old). The highest increase in sensitivity was on the hand/wrist region (+18.33 points, 95% CI: [10.74, 25.93], p<.001).

Conclusion:
The aid of AI increased sensitivity by an average of 10% without a significant decrease of specificity showing that AI can help reduce diagnostic errors.