Background: Splenic injury is the most common solid visceral injury in blunt abdominal trauma, and high-resolution abdominal computed tomography (CT) can adequately detect the injury. However, these lethal injuries sometimes have been overlooked in current practice. Deep learning (DL) algorithms have proven their capabilities in detecting abnormal findings in medical images. The aim of this study is to develop a three-dimensional, weakly supervised DL algorithm for detecting splenic injury on abdominal CT using a sequential localization and classification approach. Material and methods: The dataset was collected in a tertiary trauma center on 600 patients who underwent abdominal CT between 2008 and 2018, half of whom had splenic injuries. The images were split into development and test datasets at a 4 : 1 ratio. A two-step DL algorithm, including localization and classification models, was constructed to identify the splenic injury. Model performance was evaluated using the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Grad-CAM (Gradient-weighted Class Activation Mapping) heatmaps from the test set were visually assessed. To validate the algorithm, we also collected images from another hospital to serve as external validation data. Results: A total of 480 patients, 50% of whom had spleen injuries, were included in the development dataset, and the rest were included in the test dataset. All patients underwent contrast-enhanced abdominal CT in the emergency room. The automatic two-step EfficientNet model detected splenic injury with an AUROC of 0.901 (95% CI: 0.836–0.953). At the maximum Youden index, the accuracy, sensitivity, specificity, PPV, and NPV were 0.88, 0.81, 0.92, 0.91, and 0.83, respectively. The heatmap identified 96.3% of splenic injury sites in true positive cases. The algorithm achieved a sensitivity of 0.92 for detecting trauma in the external validation cohort, with an acceptable accuracy of 0.80. Conclusions: The DL model can identify splenic injury on CT, and further application in trauma scenarios is possible.

Uncontrolled post-traumatic hemorrhage is an important cause of traumatic mortality that can be avoided. This study intends to use machine learning (ML) to build an algorithm based on data collected from an electronic health record (EHR) system to predict the risk of delayed bleeding in trauma patients in the ICU. We enrolled patients with torso trauma in the surgical ICU. Demographic features, clinical presentations, and laboratory data were collected from EHR. The algorithm was designed to predict hemoglobin dropping 6 h before it happened and evaluated the performance with 10-fold cross-validation. We collected 2218 cases from 2008 to 2018 in a trauma center. There were 1036 (46.7%) patients with positive hemorrhage events during their ICU stay. Two machine learning algorithms were used to predict ongoing hemorrhage events. The logistic model tree (LMT) and the random forest algorithm achieved an area under the curve (AUC) of 0.816 and 0.809, respectively. In this study, we presented the ML model using demographics, vital signs, and lab data, promising results in predicting delayed bleeding risk in torso trauma patients. Our study also showed the possibility of an early warning system alerting ICU staff that trauma patients need re-evaluation or further survey.