Deep-ASPECTS: A Segmentation-Assisted Model for Stroke Severity Measurement

Ujjwal Upadhyay

Abstract: A stroke occurs when an artery in the brain ruptures and bleeds or when the blood supply to the brain is obstructed, preventing blood and oxygen from reaching brain tissue and resulting in cell death. The Middle Cerebral Artery (MCA), the largest cerebral artery, is the vessel most commonly affected in stroke. The sudden onset of a focal neurological deficit caused by interruption of blood flow within the MCA territory is referred to as an MCA stroke. The Alberta Stroke Programme Early CT Score (ASPECTS) is widely used to estimate the extent of early ischemic changes in patients with MCA stroke. In this study, we propose a deep learning–based approach to automatically score CT scans for ASPECTS evaluation. Our work has three main contributions. First, we introduce a novel medical image segmentation framework for stroke detection. Second, we demonstrate the effectiveness of an AI-driven solution for fully automated ASPECT scoring, significantly reducing diagnosis time for a given non-contrast CT (NCCT) scan. Third, we show that our model achieves a Dice Similarity Coefficient of $0.64$ for MCA anatomy segmentation and $0.72$ for infarct segmentation, with overall performance comparable to the inter-reader variability observed among expert radiologists.