BriXNet: Efficient Depthwise-Separable CNN for Lightweight Brain Stroke Classification from CT Images

• We design a compact architecture containing only 27.91K
model parameters, built primarily from depthwiseseparable convolutions.
• We evaluate the proposed model on two public CT-image
datasets using accuracy, precision, sensitivity, specificity,
F1-score, Matthews correlation coefficient (MCC), and
AUC.
• We conduct within-dataset, cross-dataset, and five-fold
analyses, with explicit discussion of data leakage and
external-validity risks