AI model detects femoral-neck fractures on hip and pelvic radiographs with high sensitivity and faster reader interpretation

Deep learning model OccuNet detected femoral-neck fractures on hip/pelvic radiographs with 97.5% sensitivity, 98.8% specificity, AUC 0.99. It raised reader sensitivity and cut reading time: radiologists 93.7% to 97.2%, emergency physicians 84.3% to 95.6%, time -14.9% and -18.9%.

• For radiograph-negative or indeterminate femoral-neck fractures (n = 189), OccuNet achieved 94.7% sensitivity, higher than musculoskeletal radiologists (86.2%) and emergency medicine physicians (68.8%; both P < .001).
• The multicenter retrospective study included 2576 patients (mean age 69 years; 1380 women) across four hospitals, with internal and three external test sets.
• With artificial intelligence assistance, radiologist sensitivity rose from 93.7% to 97.2% and emergency medicine physician sensitivity from 84.3% to 95.6%, while reading times fell 14.9% and 18.9% (all P < .001).

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.