Deep Learning-Based Aortic Diameter Measurement in Traumatic Hemorrhage Using Shallow Attention Network: A Path Forward

Background/Objectives: The accurate assessment of aortic diameter (AoD) is essential in managing patients with traumatic hemorrhage, particularly during interventions such as resuscitative endovascular balloon occlusion of the aorta (REBOA). Manual AoD measurements are time-consuming and subject to inter-observer variability. This study aimed to develop and validate a deep learning (DL) model for automated AoD measurement in trauma patients requiring massive transfusion. Methods: Abdominal CT scans from 300 adult patients were retrospectively analyzed. A Shallow Attention Network was trained on 444 manually annotated axial CT images to segment the aorta and measure its diameter. An ellipse-based calibration method was employed for enhanced measurement accuracy. Results: The model achieved a mean Dice coefficient of 0.865 and an intersection over union of 0.9988. After calibration, the mean discrepancy between predicted and ground truth diameters was 2.11 mm. The median diaphragmatic AoD was 22.59 mm (interquartile range: 20.18–24.74 mm). Conclusions: The proposed DL model with ellipse-based calibration demonstrated robust performance in automated AoD measurement and may facilitate timely planning of aortic interventions in trauma care.