MR-X pert: Hybrid optimal deep learning technique for expert-level lesion segmentation and classification of subacute ischemic stroke from MRI images

Ischemic stroke is an important neurological disease that necessitates rapid and precise diagnosis for efficient treatment. Magnetic Resonance Imaging (MRI) is frequently utilized for depicting abnormal regions within the brain since it possesses high sensitivity in detecting lesions caused by ischemia. Yet, the process of automatic segmentation and classification of such lesions is difficult owing to their complex morphology, variable sizes and poor contrast with adjacent structures. In this study, Hybrid optimal deep learning technique for expert-level lesion segmentation and classification of subacute ischemic stroke from MRI images (MR-X pert) is proposed. Firstly, it design an optimal ResNet architecture specifically tailored for semantic segmentation of FLAIR/DWI images. The hyperbolic-valued Hopfield neural network (HV-HNN) is used for the classification of subacute ischemic stroke as stroke and no-stroke. Furthermore, a new model named Y-Net, designed to achieve expert-level segmentation of subacute ischemic stroke lesions from MRI images. The framework is evaluated on the ISLES 2015 dataset, and experimental results demonstrate that the proposed approach achieves better performance compared to the existing state-of-the-art techniques in terms of Dice Similarity Coefficient and Hausdorff Distance, indicating improved segmentation accuracy and boundary refinement.