Abstract
<title>Abstract</title> <p>Background Blunt liver trauma demands rapid decisions. Non-contrast CT(NCCT) is fast but has poor soft-tissue contrast. Current models ignore clinical data for treatment prediction. We aimed to develop and evaluate a multimodal framework integrating NCCT and clinical indicators for automated identification and treatment strategy prediction. Methods This multicenter retrospective study included 103 patients with liver trauma and 200 individuals with normal liver morphology confirmed by non-contrast abdominal CT. A two-stage segmentation framework was designed, in which the liver was first delineated using model of FCN-ResNet50, followed by segmentation of traumatic regions within the liver using model of FCN-ResNet101. For treatment decision prediction, deep learning (DL) models based on non-contrast abdominal CT images and machine learning (ML) models based on clinical data were independently trained and subsequently integrated through logistic regression to construct an ensemble model. Model performance was evaluated using the Dice similarity coefficient (DSC), area under the receiver operating characteristic curve (AUC), area under the precision-recall curve (AUPRC), and additional classification metrics. Results The proposed two-stage segmentation approach achieved a DSC of 0.957 for liver delineation and 0.857 for trauma segmentation, outperforming direct whole-image segmentation (DSC = 0.840). For treatment decision prediction, the DL model based on liver-region images achieved an AUC of 0.759 (95% CI, 0.547–0.919), while the optimized AdaBoost model using nine key clinical features reached an AUC of 0.947 (95% CI, 0.855-1.000). The ensemble model integrating both imaging and clinical modalities further improved discrimination (AUC = 0.952, 95% CI, 0.859-1.000). Conclusion The proposed multimodal deep learning framework effectively localized hepatic trauma and predicted individualized treatment strategies using non-contrast abdominal CT and clinical data. This approach holds promise for assisting clinicians in the rapid evaluation and management of liver trauma, improving diagnostic efficiency, and advancing personalized care in emergency settings.</p>