StrokeNeXt: A Siamese-encoder Approach for Brain Stroke Classification in Computed Tomography Imagery
Researchers' new dual-encoder model analyzes 6,774 CT images, distinguishing stroke types with near-perfect accuracy.
Deep Dive
Researchers Leo Thomas Ramos and Angel D. Sappa developed StrokeNeXt, a model for classifying brain strokes in 2D CT images. It uses a dual-branch design with two ConvNeXt encoders and a lightweight decoder. Evaluated on 6,774 images, it achieved up to 0.988 accuracy and F1-score, outperforming other models in detecting and classifying ischemic versus hemorrhagic strokes with statistically significant gains, low inference time, and fast convergence.
Why It Matters
This could enable faster, more accurate emergency stroke diagnosis from standard CT scans, directly impacting critical treatment decisions.