Assessment of Rib Fracture in Acute Trauma Using Automatic Rib Segmentation and a Curved, Unfolded View of the Ribs: Is There a Saving of Time?

Introduction: The fast and accurate diagnosis of rib fractures in polytrauma patients is important to reduce the mortality rate and relieve long-term pain and complications. Aim: To evaluate the diagnostic accuracy and potential time savings using automatic rib segmentation and a curved, unfolded view for the detection of rib fractures in trauma patients. Methods: The multidetector computed tomography raw data of 101 consecutive polytrauma patients (72 men; mean age 45 years, age range 17 to 84 years) admitted to a university hospital were retrospectively post-processed to generate a curved, unfolded view of the rib cage. No manual corrections were performed. Patients with reconstruction errors and movement artifacts were excluded from further analysis. All fractures were identified and classified by the study coordinator using the original data set. Two readers (reader 1 and reader 2) evaluated the original axial sections and the unfolded view, separately. The fracture locations, fracture type, and reading times were recorded. Sensitivity and specificity were calculated on a per-rib basis using a ratio estimator. Cohen's Kappa was calculated as an index of inter-rater agreement. Results: 26 of 101 patients (25.7%) were excluded from further analysis owing to breathing artifacts (6.9%) or incorrect centerline computation in the unfolded view (18.8%). In total, 107 (5.9%) of 1800 ribs were fractured in 25 (33%) of 75 patients. The unfolded view had a sensitivity/specificity of 81%/100% (reader 1) and 71%/100% (reader 2) compared to 94%/100% (reader 1; p = 0.002/p = 0.754) and 63%/99% (reader 2; p < 0.001/p = 0.002). The sensitivity (reader 1; reader 2) was poor for buckled fractures (31%; 38%), moderate for undislocated fractures (78%; 62%), and good for dislocated fractures (94%; 90%). The assessment of the unfolded view was performed significantly faster than that of the original layers (19.5 ± 9.4 s vs. 68.6 ± 32.4 s by reader 1 (p < 0.001); 24.1 ± 9.5 s vs. 40.2 ± 12.7 s by reader 2 (p < 0.001)). Both readers demonstrated a very high interobserver agreement for the unfolded view (κ = 0.839) but only a moderate agreement for the original view (κ = 0.529). Conclusion: Apart from a relatively high number of incorrect centerline reconstructions, the unfolded view of the rib cage allows a faster diagnosis of dislocated rib fractures.