Brain Stroke Prediction and Classification

Abstract

A A stroke is a serious illness that can belife-threatening.Itoccurswhenbloodflowtothebrain is interrupted, which can cause cell death or damage. To lower death rates and avoid irreversible brain damage, early and precise stroke detection is essential. Conven-tionaldiagnosistechniquesmainlydependonradiologists, which can be laborious and prone to human error, particularlyinemergencysituations.Anautomatedsystem for classifying brain strokes from CT scan images using Deep Learning techniques is presented in this project. To categorisephotosintotwogroups—StrokeandNormal—a Convolutional Neural Network (CNN) model is created.In order to improve robustness and prevent overfitting, the model is trained using pre-processed medical imaging data with augmentation. Accuracy, validation accuracy, and confusion matrix metrics are used to evaluate perfor-mance.

Introduction

Stroke is a serious neurological disorder caused by interruption of blood flow to the brain, leading to brain cell damage, disability, and death. Early and accurate stroke diagnosis is essential because timely treatment improves patient recovery. CT imaging is widely used for stroke detection, but manual interpretation is time-consuming and may be affected by radiologist experience and fatigue. To overcome these limitations, Artificial Intelligence (AI), especially Deep Learning models such as Convolutional Neural Networks (CNNs), are being used to automatically analyze brain CT images and identify stroke-related patterns.

The proposed system focuses on developing an intelligent deep learning-based stroke detection model that classifies CT scans into normal and stroke categories. CNN models automatically extract important image features and provide accurate predictions, making them useful as clinical decision-support tools, especially in remote healthcare environments.

Traditional stroke detection methods used handcrafted features with machine learning algorithms such as Support Vector Machines, Random Forests, and Naïve Bayes. However, these approaches required manual feature extraction and had limited performance due to imaging variations. CNN-based approaches improved stroke detection by automatically learning complex features from medical images. Transfer learning models such as VGG16, ResNet, and DenseNet have also been explored, while Explainable AI (XAI) techniques like Grad-CAM, LIME, and SHAP improve model transparency by showing brain regions influencing predictions.

The methodology includes collecting a dataset of approximately 1,200 brain CT images, preprocessing them through resizing, normalization, and augmentation, and training a CNN model for classification. The system evaluates performance using accuracy, precision, recall, and F1-score. The developed model achieved 96.2% accuracy, 95.4% precision, 96.0% recall, and 95.7% F1-score, demonstrating strong capability in distinguishing stroke and normal CT images.

Data augmentation methods such as rotation, flipping, and zooming improved model generalization and reduced overfitting. The CNN model was compared with other architectures such as VGG16, ResNet50, and DenseNet121, where CNN achieved the best balance of accuracy, speed, and computational efficiency, making it more suitable for real-time clinical applications.

Overall, the study shows that combining medical imaging with deep learning can create reliable automated stroke detection systems. CNN-based models can assist doctors by improving diagnostic speed, accuracy, and accessibility while reducing the workload on radiologists. Future improvements include larger datasets, better explainability, and real-world clinical validation for safer AI-based stroke diagnosis.

Conclusion

TohelpwiththeearlydetectionofstrokeusingCTimages, this study introduces an AI-based Brain Stroke Classification system. By using deep learning and image processing tech-niques,theproposedConvolutionalNeuralNetwork(CNN) effectively and accurately distinguishes between stroke and normalcases.Thesystemaddressesseveralimportantclinical challenges, such as poor manual interpretation, delayed diag-nosis, and limited access to skilled radiologists. Especially in healthcaresettingswithfewresources,itsautomatedandreal-time prediction capability supports timely medical decision-making and boosts diagnostic confidence. Additionally, the model’s lightweight design ensures fast inference with low computational needs, making it easy to integrate into existing clinical workflows and telemedicine platforms.

References

[1] S. Chilamkurthy,R.Ghosh, S.Tanamala,M.Biviji, N.G.Campeau, V.K. Venugopal, V. Mahajan, P. Rao, and P. Warier, “Development andvalidation of deep learning algorithms for detection of critical findingsin head CT scans,” arXiv preprint arXiv:1803.05854, 2018. [2] RadiologicalSocietyofNorthAmerica(RSNA),“RSNA2019BrainCTIntracranial Hemorrhage Detection Challenge,” 2019. [Online]. Avail-able:https://www.kaggle.com/c/rsna-intracranial-hemorrhage-detection [3] M. Burduja, A. Naeini, and M. Al-Sharman, “Accurate and efficientintracranial hemorrhage detection on head CT using deep learning,”Radiology: Artificial Intelligence, vol. 3, no. 6, p. e210102, 2021. [4] X. Wang, J. Peng, and H. Lu, “Expert-level detection of acute intracra-nial hemorrhage on head computed tomography using deep learning,”Proc. Natl. Acad. Sci. U.S.A., vol. 116, no. 45, pp. 22737–22746, 2021. [5] A. Kundisch, J. S. R. Smith, and T. Zimmermann, “Deep learningalgorithm in detecting intracranial hemorrhages: evaluation of errorsand performance,” PLOS ONE, vol. 16, no. 5, p. e0251485, 2021. [6] J. Y. Lee, H. Kim, and Y. Kim, “Detection and classification ofintracranialhemorrhageonCTusingdeeplearning,”ScientificReports,vol. 10, no. 1, p. 20546, 2020. [7] P. Inkeaw, P. Sanpavat, and P. Tongdee, “Automatic hemorrhage seg-mentation on head CT scan for subtype classification using deeplearning,” Comput. Med. Imaging Graph., vol. 100, p. 102097, 2022. [8] O. Ozaltin and H. U. Yildiz, “A deep learning approach for detectingstroke from brain CT images: OzNet,” Bioengineering, vol. 9, no. 12, p.783,2022,doi:10.3390/bioengineering9120783. [9] L.Corte´s-Ferre,A.Torres,andJ.S.Rodr´?guez,“Deeplearningappliedto intracranial hemorrhage detection: review and perspectives,” Front.Comput. Neurosci., vol. 17, p. 1152782, 2023. [10] M. J. Ferdous, “An ensemble CNN model for brain stroke predictionfromCTscans(ENSNET),”ProcediaComput.Sci.,vol.235,pp.112–121, 2024. [11] T. D’Angelo, M. Raftopoulos, and M. F. Bottari, “Accuracy and timeefficiency of a novel deep learning algorithm for identification andclassificationofintracranialhemorrhagefromunenhancedCT,”Insightsinto Imaging, vol. 15, no. 1, p. 105, 2024. [12] J.N.D.Fernandes,R.H.Sousa,andM.T.Oliveira,“Machineanddeeplearning in brain stroke diagnosis: a comprehensive review,” Sensors,vol. 24, no. 4, p. 1358, 2024. [13] F.Shang,S.Wang,X.Wang,andY.Yang,“Transformer-basedapproachforintracranialhemorrhagedetection,”Proc.AAAIWorkshoponAIforMedicine, 2022. [14] A. Karamian, L. P. Rossi, and F. Chen, “Diagnostic accuracy of deeplearning for intracranial hemorrhage: a meta-analysis,” J. Clin. Med.,vol. 14, no. 3, p. 755, 2025. [15] S. Qari, “Brain stroke detection and classification using CT scans withCNNandsegmentationapproaches,”arXivpreprintarXiv:2507.09630,2025. [16] E. Ekingen, M. A. Aydin, and H. Sezer, “StrokeNeXt: an automatedstroke classification model using lightweight CNN for CT and MR,”BMC Med. Imaging, vol. 25, no. 1, p. 48, 2025. [17] H. Abdi, M. U. Sattar, R. Hasan, V. Dattana, and S. Mahmood, “Strokedetection in brain CT images using convolutional neural networks:modeldevelopment,optimizationandinterpretability,”Information,vol.16, no. 5, p. 345, 2025. [18] S.Author,M.Shah,andA.Patel,“Accuracyofcombineddeeplearningalgorithms in detecting intracranial hemorrhage subtypes from non-contrast CT,” medRxiv preprint, May 2024.

Copyright

Copyright © 2026 Shrikanth N G, Vinayak M Naganur, Sanjeev Reddy, Mithun Gowda SR. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.