Ischemic stroke, a cerebrovascular disorder characterized by obstructions in cerebral blood vessels, necessitates accurate segmentation of brain lesions for effective diagnosis, treatment, and rehabilitation in clinical practice. Magnetic Resonance Imaging is the preferred modality for stroke imaging due to its superior soft tissue contrast and versatility. Despite this, manual segmentation of stroke lesions from MRI is labor-intensive and demands considerable expertise, highlighting the need for automated solutions. The complexity of ischemic lesions, manifested through variability in size and location, poses significant challenges for automated segmentation. Recently, nnU-Net has emerged as a state-of-the-art framework for medical image segmentation, recognized for its self-configuring capability across diverse datasets. However, as deep learning research evolves and new architectures emerge, nnU-Net’s inherent complexity may hinder its ability to keep pace with cutting-edge advancements, potentially impeding it from realizing its full potential. To address these limitations, we propose a novel approach incorporating the 3D-EfficientConv block, optimized for volumetric data to enhance feature extraction and spatial relationships, coupled with the 3D block attention module to refine feature representation. We customized nnU-Net by incorporating the 3D-EfficientConv block as a dynamic repeatable core unit while preserving nnU-Net’s robust preprocessing pipeline, thus leveraging the strengths of both. Our approach demonstrates promising results, surpassing existing nnU-Net configurations with a Dice coefficient of 0.86 and precision of 0.87 on the ATLAS v2.0 dataset.