High-Accuracy 3-Class Cerebral Stroke Detection Using ConvNeXt: An End-to-End Vision Pipeline

DOI

10.2991/978-94-6239-664-7_11How to use a DOI?

Keywords

Cerebral Stroke; ConvNeXt; CT Imaging; Deep Learning; Data Augmentation

Abstract

Cerebral stroke is a leading global cause of death and long-term disability. Early, accurate identification of stroke presence and subtype (ischemic vs hemorrhagic) from brain imaging is essential because therapies differ widely (e.g., thrombolysis for ischemic vs surgery in hemorrhagic). Nevertheless, manual CT interpretation is time-consuming and observer dependent, especially in under-resourced settings. This work targets automatic high-accuracy classification to aid medical diagnosis. An end-to-end ConvNeXt-based pipeline is designed for multi-class stroke classification with aggressive data augmentation to address class imbalance and improve generalization. It outperforms established baselines (ResNet-50, DenseNet-121, VGG-16, and VGG-19) in comprehensive evaluation. Ablation studies and error analysis show how architectural choices and extensions affect performance. We fine-tune a ConvNeXt-Tiny model on grayscale CT images resized to 224×224 and replicated to RGB, with augmentations such as rotations, distortions, and noise. Training uses AdamW, cosine-annealing learning rate scheduling, and cross-entropy loss. On the TEKNOFEST 2021 Stroke Dataset (~12,000 heavily imbalanced CT scans), our model surpasses ResNet-50 by 0.85% accuracy and 1.75% F1, reaching 97.02% test accuracy, 95.84% macro precision, 96.02% macro recall, and 95.93% macro F1-score. This study advances AI-assisted stroke diagnosis in emergency care, enabling faster triage and potentially reducing mortality through accurate, deployable models in practice.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Amit Kumar Ghosh
AU  - Mst Happy Akther
AU  - Shahriar Marjan
AU  - Md. Monarul Islam Mithu
PY  - 2026
DA  - 2026/06/08
TI  - High-Accuracy 3-Class Cerebral Stroke Detection Using ConvNeXt: An End-to-End Vision Pipeline
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 133
EP  - 144
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_11
DO  - 10.2991/978-94-6239-664-7_11
ID  - Ghosh2026
ER  -