Angel D. Sappa

D convolutional fusion path with transformation layers, and a classification head. • We present StrokeNeXt, an approach for brain stroke classification that integrates dual-branch feature extraction and a lightweight decoder. • We demonstrate that StrokeNeXt achieves high accuracy in distinguishing between stroke and normal cases using real-world CT scan data. • We show that the model also performs effectively in differentiating among various stroke subtypes. • We show that StrokeNeXt performs notably when compared to other state-of-the-art models from the literature. • We establish a baseline architecture and performance benchmark that can serve as reference for future research. 2 Related work Stroke identification has been widely explored using diverse paradigms. Early approaches relied on traditional classifiers such as Random Forest [4] and ElasticNet [8], which achieved moderate success in stroke detection but were limited by their dependence on handcrafted features and lack of spatial context. Subsequent efforts shifted toward DL, where CNN-based models such as MobileNetV2 [29] improved automation and spatial feature extraction but still exhibited limited representational capacity and robustness. Other architectures have been proposed to enhance feature representation. Models such as P-CNN [11], D-UNet [29], and OzNet [20] improve lesion localization through multi-scale encoding or task-specific optimizations. In parallel, D Computed Tomography (CT) images. StrokeNeXt employs a dual-branch design with two ConvNeXt encoders, whose features are fused through a lightweight convolutional decoder based on stacked, D pipelines, increasing complexity and overfitting risk, limiting their suitability for real-time clinical deployment. Recent work has explored hybrid architectures to model long-range dependencies. For example, StrokeViT [21] combines convolutional networks with Vision Transformers to enhance feature representation and prediction accuracy. However, Transformer-based models generally demand large datasets, high computational cost, and long training times, limiting their practicality in resource- and time-constrained clinical settings. Overall, existing methods exhibit a trade-off between accuracy, efficiency, and reliability. While some achieve strong classification performance, they often incur high computational cost, whereas others lack proper calibration for time-sensitive clinical deployment. Addressing these limitations, StrokeNeXt proposes a dual-branch convolutional framework that strengthens feature diversity through parallel processing while maintaining efficiency. Its design focuses on achieving robust diagnostic performance without relying on large data volumes or computationally expensive inference pipelines, positioning it as a balanced alternative within the current landscape of stroke analysis methods. 3 Methodology 3.1 Dataset and preprocessing We use the Stroke Dataset released as part of the Artificial Intelligence in Healthcare Competition (TEKNOFEST 2021) [15]. It consists of 6,774 brain CT cross-sectional images in PNG format, annotated and curated by a team of seven radiologists. The dataset is divided into three classes: 4,551 non-stroke, 1,093 hemorrhagic stroke, and 1,130 ischemic stroke cases. The dataset is publicly available, and representative samples are shown in Fig. 2. Figure 2: Sample images from the used dataset in this work. To structure the experiments, the dataset was reorganized into two settings. The first addresses stroke detection by grouping ischemic and hemorrhagic cases into a single positive class and using non-stroke samples as negatives. The second focuses on stroke subtype classification, considering only ischemic and hemorrhagic cases and excluding non-stroke images. In both settings, the data were randomly split into training, validation, and test sets following an 80-10-10 ratio. Data augmentation was used, including random horizontal flips, rotations up to 10°, and brightness, contrast, and saturation jittering. Images were resized to