Detection and Localization of Subdural Hematoma Using Deep Learning on Computed Tomography

TL;DR

This study tackles rapid, accurate SDH detection and localization by integrating clinical data, volumetric CT analysis, and pixel-level segmentation into a single multimodal pipeline. By combining a tabular XGBoost model, a 3D CNN on CT volumes, and a transformer-enhanced 2D segmentation network within an ensemble, the approach achieves high discrimination (AUC ≈0.941) and anatomically meaningful localization. The segmentation component yields robust spatial maps, and the system demonstrates strong cross-domain generalization, with potential to improve triage, reduce time to intervention, and standardize SDH management in clinical workflows. External validation and larger annotated segmentation datasets are planned to support prospective deployment in EMR-integrated radiology workflows.

Abstract

Background. Subdural hematoma (SDH) is a common neurosurgical emergency, with increasing incidence in aging populations. Rapid and accurate identification is essential to guide timely intervention, yet existing automated tools focus primarily on detection and provide limited interpretability or spatial localization. There remains a need for transparent, high-performing systems that integrate multimodal clinical and imaging information to support real-time decision-making. Methods. We developed a multimodal deep-learning framework that integrates structured clinical variables, a 3D convolutional neural network trained on CT volumes, and a transformer-enhanced 2D segmentation model for SDH detection and localization. Using 25,315 head CT studies from Hartford HealthCare (2015--2024), of which 3,774 (14.9\%) contained clinician-confirmed SDH, tabular models were trained on demographics, comorbidities, medications, and laboratory results. Imaging models were trained to detect SDH and generate voxel-level probability maps. A greedy ensemble strategy combined complementary predictors. Findings. Clinical variables alone provided modest discriminatory power (AUC 0.75). Convolutional models trained on CT volumes and segmentation-derived maps achieved substantially higher accuracy (AUCs 0.922 and 0.926). The multimodal ensemble integrating all components achieved the best overall performance (AUC 0.9407; 95\% CI, 0.930--0.951) and produced anatomically meaningful localization maps consistent with known SDH patterns. Interpretation. This multimodal, interpretable framework provides rapid and accurate SDH detection and localization, achieving high detection performance and offering transparent, anatomically grounded outputs. Integration into radiology workflows could streamline triage, reduce time to intervention, and improve consistency in SDH management.

Figures (3)

• Figure 1: Overview of the proposed multimodal pipeline for subdural hematoma (SDH) detection and localization. The system integrates three complementary components: a tabular XGBoost model using clinical variables, an ensemble of 3D convolutional neural networks (CNNs) applied to CT scans, and a 2D Swin-CNN hybrid segmentation model for hematoma localization. Probabilities from each component are combined into a final ensemble output indicating the likelihood of SDH presence.
• Figure 2: Model evaluation curves on the test set. (a) ROC curves demonstrate strong discrimination (AUCs above 0·90 across all deep learning model groups). (b) Calibration curves show well-aligned predicted and observed probabilities, supporting reliability for clinical use.
• Figure 3: Examples of subdural hematoma (SDH) localization test set results from the proposed segmentation model.