Contextual transformer for brain stroke segmentation in MRI

Accurate segmentation of brain stroke using magnetic resonance imaging (MRI) is associated with difficulties due to the complicated anatomy of the brain and the different properties of the stroke. A well-known challenge associated with this problem is the variability in the size and shape of the stroke, which makes it difficult to employ standard segmentation strategies. This study introduces a contextual transformer framework, which synergies spatial feature extraction with global anatomical processing ability, further enhanced by advanced feature fusion and segmentation synthesis techniques. The main idea is based on focus in and generalize out (FIGO), which uses a special pipeline that focus in stroke features with generalization out external anatomical structures. Experimental studies using the ATLAS v2.0 dataset demonstrate a remarkable improvement in segmentation accuracy using FIGO. Ablation studies confirm the significant impact of the proposed framework in comparison with conventional methods. • Brain stroke segmentation suffers from several challenges due to the complexity of brain anatomy and lesion variability. • Focus In and Generalize Out (FIGO) framework is a novel brain stroke segmentation that efficiently blends global and local features. • Experimental studies using the ATLAS v2.0 dataset demonstrate a remarkable improvement in segmentation accuracy. • The proposed framework enables better observation of strokes with various shapes, sizes, and locations.