Development and Evaluation of a Deep Learning Algorithm for Rib Segmentation and Fracture Detection from Multicenter...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Rib fractures are common injuries often diagnosed using computed tomography (CT).
• Accurate and timely detection of rib fractures is crucial for patient management and outcomes.
• Current diagnostic methods rely heavily on radiologist interpretation, which can be time-consuming and prone to inter-observer variability.

Purpose of the Study:

• To assess the efficacy of a deep learning (DL) algorithm for automated detection and localization of rib fractures on chest CT images.
• To compare the performance of the DL algorithm against human radiologists in identifying rib fractures.
• To evaluate the impact of the DL algorithm as an assistive tool for radiologists.

Main Methods:

• A retrospective analysis of 10,943 chest CT scans from six hospitals was conducted.
• A DL algorithm was trained and validated on distinct datasets for fracture detection and rib segmentation.
• Performance metrics included Free-response receiver operating characteristic (FROC) score, precision, sensitivity, F1 score, Area Under the Curve (AUC), Dice coefficient, and accuracy.

Main Results:

• The DL algorithm achieved an FROC score of 84.3% for rib fracture detection.
• On a separate test set, the algorithm demonstrated detection performance (precision 82.2%, sensitivity 84.9%, F1 score 83.3%) comparable to radiologists.
• When used adjunctively, the algorithm improved radiologist sensitivity from 79.7% to 89.2% while maintaining high precision.
• The model achieved an AUC of 0.93 for classification and a Dice score of 0.827 for rib segmentation.

Conclusions:

• The developed deep learning algorithm effectively detects rib fractures and their anatomical locations on CT images.
• The algorithm shows potential as a valuable tool to assist radiologists in diagnosing rib fractures.
• This technology could enhance diagnostic accuracy and efficiency in clinical practice.