Abstract

This study presents a comprehensive framework that integrates a deep learning model with advanced image preprocessing techniques to improve the multilabel classification of five types of intracranial hemorrhage—epidural, intraparenchymal, intraventricular, subarachnoid, and subdural—using non-contrast computed tomography (CT) images. The framework includes strategies to mitigate overfitting, data augmentation, and a custom loss function. It was rigorously evaluated on a dataset of over 25,000 non-contrast CT scans, each labeled by expert radiologists across six classes. The proposed model achieved 99% accuracy and 92.1% sensitivity in detecting intracranial hemorrhage, outperforming previously reported methods. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.