EUVIMEDEUVIMED entdecken
Crossref

Deep Learning-Based Brain Stroke Detection from CT Scan Images

Harshal Mali Harshal Mali

International Journal of Scientific Research in Engineering and Management · 2026-06 · 10 (6)

DOI: 10.55041/ijsrem65044 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Scientific Research in Engineering and Management, 2026-06. Verfasst von Harshal Mali Harshal Mali.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Brain stroke is among the foremost causes of death and long-term disability worldwide, with the World Health Organization reporting approximately 15 million strokes annually, resulting in five million deaths and a further five million permanent disabilities.Rapid and accurate diagnosis from computed tomography (CT) imaging is critical in emergency settings; however, manual CT interpretation remains prone to inter-observer variability and missed findings under time pressure.This paper proposes a hybrid deep learning framework -EfficientNet-B0 combined with a Bidirectional Long Short-Term Memory (BiLSTM) network -for automated three-class brain stroke classification from CT scans, targeting Normal, Bleeding (Hemorrhagic), and Ischemia categories.The EfficientNet-B0 backbone, pretrained on ImageNet, extracts 1280-dimensional spatial feature maps subsequently compressed to 256 dimensions via a 1×1 convolutional layer.These features are reshaped into a spatial sequence and modelled by a BiLSTM network (hidden size 128, bidirectional), which captures spatial-temporal contextual relationships across feature map positions.Trained on 6,650 CT images with a class-weighted cross-entropy loss and the Adam optimiser, the model achieves 97.14% test accuracy, a macro precision of 96.94%, macro recall of 95.47%, and macro F1-score of 96.18%, outperforming Simple CNN (86.90%) and ResNet-50 (92.57%) baselines by significant margins.The system is deployed as a full-stack web application (ReactJS + Flask + PyTorch) supporting real-time CT upload and diagnostic report generation.Experimental results confirm that combining spatial feature extraction with temporal sequence modelling substantially improves stroke classification performance, offering a reliable clinical decision support tool for emergency radiology.

Zitieren

APA

Mali, H. M. H. (2026). Deep Learning-Based Brain Stroke Detection from CT Scan Images. International Journal of Scientific Research in Engineering and Management, 10(6). https://doi.org/10.55041/ijsrem65044

Vancouver

Mali HMH. Deep Learning-Based Brain Stroke Detection from CT Scan Images. International Journal of Scientific Research in Engineering and Management. 2026;10(6). doi:10.55041/ijsrem65044

BibTeX
@article{Mali2026,
  author = {Mali, Harshal Mali Harshal},
  title = {Deep Learning-Based Brain Stroke Detection from CT Scan Images},
  journal = {International Journal of Scientific Research in Engineering and Management},
  year = {2026},
  volume = {10},
  number = {6},
  doi = {10.55041/ijsrem65044}
}
Kostenlos starten →