<p>Ischemic stroke is a pathological condition caused by interrupted or reduced blood flow to the brain, leading to damage or death of brain cells. Ischemic stroke continues to be a significant cause of mortality and disability worldwide. Accurate identification and localization of acute ischemic stroke (AIS) are crucial for prompt medical intervention and better patient outcomes. This study introduces a novel deep neural network (DNN) called JONet, which is a directed acyclic graph (DAG) network that incorporates a multi-head attention layer to accurately detect AIS in magnetic resonance images (MRI) using the Anatomical Tracing of Lesions After Stroke (ATLAS) dataset. The proposed framework integrates an image fusion method, and JONet incorporates multi-attention layers to effectively prioritize important features in MRI data. Moreover, the research employs generative adversarial networks (GAN) for data augmentation, thus enhancing the deep learning process by augmenting medical data. Pre-processing included selecting the middle slice and applying a quadtree fusion technique with contrast-limited adaptive histogram equalization (CLAHE) and anisotropic filtering to improve the image quality. The classification accuracy for AIS identification in the present study was 99.25%. In this study, the performance of the proposed architecture was compared with that of the current edge deep learning architecture. The results demonstrate an improvement in performance over existing architectures, making it a reliable tool for clinical practice predictions concerning brain abnormalities.</p>