MRI acute/sub-acute ischemic stroke segmentation with deep learning: A comprehensive review

The segmentation of ischemic stroke lesions from Magnetic Resonance Imaging (MRI) images using deep learning (DL) techniques has emerged as a critical area of research in medical imaging. This article provides a comprehensive review of the current state-of-the-art methodologies in this domain, focusing on the advancements and challenges inherent in this field. Our review focuses on studies that utilize DL models for segmenting acute and sub-acute ischemic stroke lesions using MRI modalities. By systematically analyzing research from 2020 onward, we aim to clarify the advancements in the published models' effectiveness and provide a performance benchmark. This review stands out by comprehensively analyzing all the studies in this field, beyond the scope of prior reviews focused only on key publications. Additionally, this work serves as a comprehensive reference for researchers by compiling all relevant datasets, MRI modalities, evaluation metrics, loss functions, input data dimensionality, preprocessing, and augmentation techniques employed for this task. Additionally, we identify the challenges in this field and highlight the existing research gaps, in addition to proposing some directions for future work to enhance the effectiveness and accuracy of DL models in stroke lesion segmentation.