OccuNet AI detects 94.7% of initially missed femoral neck fractures

AI model OccuNet detected 94.7% of femoral neck fractures initially missed or deemed indeterminate on radiographs, compared with 86.2% for musculoskeletal radiologists and 68.8% for emergency physicians. Overall sensitivity 97.5%, specificity 98.8%.

• AI assistance reduced radiologist reading time by 14.9% and emergency physician reading time by 18.9%.
• The model was developed with a two-stage contrastive pre-training approach to enhance sensitivity to subtle fractures.

Automated summary

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