Prediction of Brain Stroke Risk Using Machine Learning and Deep Learning: An Hybrid XGBoost-ResNet-18 CDSS

Abstract

Stroke continues to be one of the major causes of global deaths and neurologic impairment. Conventional clinical scoring system such as CHADS? scores static and additive in nature and unable to process high dimensional multi-modal data. This paper proposes a hybrid artificial intelligence (AI) based clinical decision support system (CDSS) for evaluating brain stroke risk based on a combination of structured clinical indicators with medical images. The hybrid model comprises of an XGBoost classifier trained on tabular biomarkers such as hypertension status, average glucose level and body mass index (BMI), and a ResNet-18 deep learning model using transfer learning to recognize stroke indicators from MRI/CT scans. With a view to addressing the black-box nature inherent in medical AI applications, we incorporate Explainable AI (XAI) techniques such as SHAP (SHapley Additive exPlanations) to provide an interpretation at the feature level for each prediction. Our prototype utilizes a decoupled React 18 – Flask architecture with PostgresSQL database back end, JSON Web Token (JWT) based role-based authentication and a forensics traceable audit trail with automatic generation of clinical PDF reports.

Introduction

The text presents an AI-powered Brain Stroke Risk Prediction and Clinical Decision Support System (CDSS) that combines clinical patient data and MRI/CT brain images to provide an integrated stroke-risk assessment.

1. Introduction

Traditional stroke-risk assessment relies on manual scoring systems, which may not capture complex relationships between multiple risk factors. The proposed system uses Machine Learning (ML) to analyze factors such as age, hypertension, glucose level, BMI, smoking, and cardiovascular conditions.

For medical imaging, Deep Learning (DL), specifically ResNet-18 with transfer learning, is used to identify stroke-related indicators in MRI/CT scans. The system also incorporates SHAP (SHapley Additive exPlanations) to explain why the AI produces a particular prediction, helping improve clinical transparency.

2. Literature Survey

Previous approaches evolved from simple linear scoring systems such as CHADS? to advanced ML models such as SVM and XGBoost. XGBoost is highlighted for its ability to handle complex relationships, imbalanced datasets, missing values, and regularization.

CNNs and residual networks such as ResNet-18 have improved medical-image analysis, particularly when datasets are too small to train deep networks from scratch. However, the lack of interpretability remains a major concern, which motivates the use of SHAP.

Recent research also supports multimodal fusion, where clinical/tabular information and medical images are combined to provide a more comprehensive prediction than either data type alone.

3. Existing vs. Proposed System

The existing approach mainly depends on manual scoring systems such as CHADS? and NIHSS, which can be subjective and have difficulty representing nonlinear interactions among risk factors.

The proposed system instead provides:

• Combined analysis of clinical data and brain images.

• A single risk probability/index.

• SHAP-based explanations for individual predictions.

• Secure access using JWT authentication and role-based access control.

• Automated clinical report generation.

4. System Architecture and Methodology

The system follows a React–Flask–PostgreSQL architecture:

• Frontend: React 18 provides doctor and administrator dashboards, form validation, predictions, and SHAP visualizations.

• Backend: Flask handles requests, preprocessing, and model inference.

• Database: PostgreSQL stores patient information, predictions, and audit logs.

• Security: JWT authentication and role-based permissions distinguish doctors from administrators

Conclusion

In summary, this paper has described an AI-Powered Brain Stroke Risk Prediction and Clinical Decision Support System which integrates an XGBoost classifier of structured clinical data with a ResNet-18 deep neural network for medical image analysis into a single Risk Probability Index. With the incorporation of Explainable AI based on SHAP values, this approach directly tackles the challenge of the “black box” problem, providing a feature-level explanation for each prediction instead of the probability value. From the architectural perspective, the separation of React 18 frontend and Flask backend, the PostgreSQL database, the use of JWT-based Role-Based Access Control and forensic auditing, and the creation of PDF reports prove that the system was designed not just as a research prototype but as a potential candidate for implementation in clinical practice. Thus, this paper demonstrates that the ability to make predictions can be combined with explainability and audibility in one system with the aim to help detect stroke risk earlier and more evidence-based.

References

[1] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778. [2] T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, 2016, pp. 785–794. [3] S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 30, 2017. [4] Kaggle, “Stroke Prediction Dataset,” 2021. [Online]. Available: https://www.kaggle.com/datasets/fedesoriano/stroke-prediction-dataset [5] A. Paszke et al., “PyTorch: An imperative style, high-performance deep learning library,” in Advances in Neural Information Processing Systems (NeurIPS), vol. 32, 2019. [6] M. Grinberg, Flask Web Development: Developing Web Applications with Python. Sebastopol, CA: O’Reilly Media, 2018. [7] A. Boduch and R. Derks, React and React Native: A Complete Guide to Building Cross-Platform Apps with React. Birmingham, U.K.: Packt Publishing, 2020. [8] World Health Organization, “Stroke: A global medical emergency,” 2022. [Online]. Available: https://www.who.int [9] M. Winand, SQL Performance Explained. Self-published, 2012. [10] L. O. Gostin and S. J. Nass, Beyond the HIPAA Privacy Rule: Enhancing Privacy, Improving Health Through Research. Washington, DC: National Academies Press, 2009. [11] ReportLab Europe Ltd., “ReportLab PDF Library User Guide,” 2024. [Online]. Available: https://www.reportlab.com/docs/reportlab-userguide.pdf [12] S. C. Huang, A. Pareek, S. Seyyedi, I. Banerjee, and M. P. Lungren, “Fusion of medical imaging and electronic health records using deep learning: A systematic review and implementation guidelines,” NPJ Digital Medicine, vol. 3, no. 1, pp. 1–9, 2020.

Copyright

Copyright © 2026 G Santhosh Raj, A Vijayan. 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.