A Deep Learning based Ischemic Stroke Lesion Segmentation using Modified Multi-Scale 3D CNN

Acute Ischemic Stroke (AIS) is the second-largest fatal accident that results in High morbidity, disability, and mortality worldwide. Clinicians can help patients receive the right therapies by doing an early and automated diagnosis of AIS. A computerized method for segmenting ischemia lesions should be developed to reduce time consumption and avoid clinical error. This paper introduces the segmentation techniques based on the Deep learning algorithm by creating an eleven-layer deep, double pathway, 3D Convolutional Neural Network. After analyzing a relevant multi-modal 3D MRI-like Diffusion Weighted Imaging (DWI) patch at various scales, the created system segments pathologies voxel-wise. For training and testing, this study uses 255 examples with five-fold validation. After postprocessing, it outperformed other approaches with a high Dice mean value of 0.82 on the datasets of ISLES 2022. The implemented network can process a 3D brain volume of different sizes within 3 minutes, making it relevant to broader research cohorts' automated analysis.