StrokeNetBench: A Comparative Framework of Deep Architectures for Stroke Detection and Classification

DOI

10.2991/978-94-6239-664-7_10How to use a DOI?

Keywords

Stroke Detection; CT Imaging; Deep Learning; VGG16; VGG19; ResNet50; ViT-Base; Focal Loss; ClassWeighting; Transfer Learning; Convolutional Neural Network; 10-Fold Cross-Validation; Grad-CAM

Abstract

Stroke is a leading cause of mortality and permanent disability worldwide. It is a serious neurological emergency. Timely intervention and better outcomes depend on accurate identification of its two major types, ischemic and hemorrhagic stroke. Manual interpretation of brain computed tomography (CT) scans is laborious, subjective, and errorprone. We provide a deep learning-based architecture for automatically classifying brain CT scans into three groups: ischemia, bleeding (hemorrhage), and no stroke (control). Our approach automatically constructs features from CT scans using convolutional neural networks (CNNs) and transfer learning. The framework incorporates preprocessing, data augmentation, and class imbalance handling using class weighting and focal loss. A custom CNN, VGG16, VGG19, ResNet50 and ViT-Base are evaluated under various layer configurations. Models are assessed using 10-fold cross-validation. ResNet50 with a single unfrozen residual block achieves the best performance, with 97.3% accuracy, 95.7% precision, 96.4% recall, 96.1% F1-score, and low false negative rates. High class separability is shown by ROC-AUC scores of 99% (No Stroke), 91% (Bleeding), and 97% (Ischemia). Grad-CAM-based visual explanations confirm the model focuses on stroke-relevant areas, enhancing interpretability. These results demonstrate deep learning models’ effectiveness for reliable stroke diagnosis, supporting clinical decision-making.

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  - Mashuka Bashar Chowdhury
AU  - Sadia Jannat Mitu
PY  - 2026
DA  - 2026/06/08
TI  - StrokeNetBench: A Comparative Framework of Deep Architectures for Stroke Detection and Classification
BT  - Proceedings of the International Conference on Intelligent Data Analysis and Applications (IDAA 2025)
PB  - Atlantis Press
SP  - 116
EP  - 132
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-664-7_10
DO  - 10.2991/978-94-6239-664-7_10
ID  - Chowdhury2026
ER  -