Automated Cerebral Vessel Segmentation Using Deep Learning for Early Detection of Cerebrovascular Diseases

Cerebrovascular diseases are a group of neurological disorders that result from weakened, blocked, or extravasated blood flow in the cerebral arteries. Although cerebrovascular diseases are the third leading cause of death worldwide, they are the primary cause of disability after the disease. Therefore, accurate cerebral vessel segmentation of Time-of-Flight Magnetic Resonance Angiographs (TOF-MRAs) data are crucial for clinical applications such as early disease diagnosis and surgical planning. This study focuses on vessel segmentation of major cerebral arteries using deep learning-based technologies, which have gained significant popularity in imaging problems and achieved unimaginable success. The study consists of two stages. In the first stage, we created a labelled dataset by combining Hessian-based filters and various image processing algorithms. In the second stage, we applied state-of-the-art architectures commonly used in segmentation problems, such as U-Net, ResUNet, ResUNet++, and TransUNet, to the vessel segmentation problem and compared their performance. Our experimental results demonstrate that the ResUNet++ architecture achieved the highest score in vessel segmentation, particularly on major cerebral arteries, with a mean intersection over union (mIoU) score of %91.6 compared to other architectures. These findings highlight the potential of deep learning-based approaches for accurate and efficient cerebral vessel segmentation in many clinical applications.