Timely surgical intervention is often critical in managing acute traumatic brain injury (TBI). Artificial intelligence (AI) may support triage by predicting neurosurgical need, especially where specialist expertise is lacking. While multimodal models that combine imaging and clinical data are often thought to improve performance, they also increase data acquisition complexity. This study evaluated whether multimodal AI models outperform unimodal models in predicting neurosurgical intervention for TBI. 3442 total patients presenting with a TBI were used in this research. Patients were split into two distinct datasets, a development dataset, and a holdout test dataset. The development dataset consisted of 2,928 adult trauma patients with CT and clinical data and was used to develop three models: clinical-only, imaging-only, and combined. Performance was assessed on the holdout test dataset of 514 patients. The clinical model achieved an AUC of 0.68 0.64–0.73, the imaging model 0.86 0.82–0.89, and the combined multimodal model 0.89 0.86–0.92. The combined model did not significantly outperform the image-only model. Given minimal gain and added complexity, image-based AI may offer a simpler, equally effective triage solution for TBI.
Smith et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: