Comprehensive Summary
This study presents a literature review on research done on deep learning (DL) models dealing with rib fractures from CT data. 25 studies were picked covering detection, segmentation and classification of rib fractures. 21 studies presented DL models for detection of rib fractures, which overall performed better than clinicians. The pooled proportion for sensitivity was 86.7% whereas that of clinicians was 75.4%. For segmentation of rib fractures, 4 studies presented DL models which had a pooled proportion of 92.4%. There was no comparison with clinicians. 10 studies presented DL models for classification which demonstrated a pooled sensitivity of 97.3% whereas that of clinicians were 88.2%, implying that the DL models performed better. This study concludes that DL models have significantly improved over the past few years and can outperform clinicians.
Outcomes and Implications
Trauma-induced rib fractures are common injuries which demand much labor to diagnosis using computed tomography (CT) data. Using AI models to process rib fracture diagnostics would divert much needed resources to other higher priority cases. This study demonstrates that the AI models performed better than clinicians in these tasks, suggesting possible use in clinical practice. Nevertheless, these studies have some limitations, such as varied results across multiple studies and limit confirmation data.