Improved performance and robustness of multi-task representation learning with consistency loss between pretexts for intracranial hemorrhage identification in head CT

The authors proposed a supervised multi-task aiding representation transfer learning network (SMART-Net) for intracranial hemorrhage (ICH) classification and segmentation using non-contrast head computed tomography (NCCT). The framework has upstream and downstream components, focusing on feature extraction and transfer learning for volume-level tasks. Experimental results from four test sets demonstrated that SMART-Net outperforms previous methods in robustness and performance for ICH classification and segmentation. The deep learning algorithms have been commercialized as part of the AVIEW NeuroCAD system by Coreline Soft, showing potential for improved emergency medical care.