Improving traumatic fracture detection on radiographs with artificial intelligence support: a multi-reader study

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Nonspecialist readers often face challenges in accurately detecting traumatic fractures on radiographs.
• The integration of artificial intelligence (AI) offers potential to enhance diagnostic performance in medical imaging.

Purpose of the Study:

• To evaluate the diagnostic performance of nonspecialist readers in detecting appendicular skeleton traumatic fractures with and without AI support.
• To assess the impact of AI on sensitivity, specificity, and interpretation time for fracture detection.

Main Methods:

• A retrospective, multi-reader, multi-case study involving 15 nonspecialist readers assessing 340 radiographic exams.
• Readers evaluated exams with and without an AI fracture detection support tool, with reading times recorded.
• Sensitivity, specificity, and false positives per patient were calculated against a reference standard established by consultant radiologists.

Main Results:

• AI support significantly improved patient-wise sensitivity (72% to 80%) and specificity (81% to 85%).
• AI led to a 29% relative reduction in missed fractures and a 21% relative reduction in false positives per patient.
• The most substantial gains were observed in detecting nonobvious fractures, with sensitivity increasing by 11 percentage points.

Conclusions:

• AI fracture detection support tools enhance the diagnostic performance of nonspecialist readers for traumatic appendicular skeleton fractures.
• The AI tool improved both sensitivity and specificity without adversely affecting interpretation time.
• This study highlights the novel application of AI in differentiating obvious and nonobvious fractures, advancing AI reader comparison studies.