The following is a summary of “Evaluating AI rib fracture detections using follow-up CT scans,” published in the October 2023 issue of Emergency Medicine by Zhou, et al.

For a study, researchers sought to compare Artificial Intelligence (AI) diagnosis of rib fractures using initial and follow-up CT scans as diagnostic criteria, evaluating the potential improvement in the detection rate of rib fractures with AI assistance.

A retrospective analysis involved 113 trauma patients who underwent both initial and follow-up CT scans. Initial and follow-up CT scans were independently considered as diagnostic criteria. AI software was employed for rib fracture detection. Three groups—Group 1 (radiologist), Group 2 (AI), and Group 3 (Radiologist with AI)—reviewed CT images, and differences in sensitivity and specificity for rib fracture diagnosis were recorded and compared.

Initial and follow-up CT scans diagnosed 589 and 712 rib fractures, respectively. Initial CT had a 17.84% missed rate, failing to detect 127 rib fractures and erroneously identifying four normal ribs as fractured. Using follow-up CT as the diagnostic standard, Group 3 demonstrated higher sensitivity (91.57%) than Group 1 (82.16%) and Group 2 (79.35%). Group 3 also exhibited higher specificity (99.70%) compared to Group 1 (99.80%) and Group 2 (84.90%). Group 3’s sensitivity was superior to both Group 1 and Group 2 (P < 0.05), while Group 2 showed lower specificity than Group 1 and Group 3 (P < 0.05).

AI-assisted diagnosis significantly enhanced rib fracture detection rates. Utilizing follow-up CT as the diagnostic standard, coupled with radiologist oversight, minimized AI misdiagnoses, emphasizing the potential of AI in improving accuracy in rib fracture assessments.

Source: sciencedirect.com/science/article/abs/pii/S0735675723003716