A Dual-Stream Deep Learning Architecture With Adaptive Random Vector Functional Link for Multi-Center Ischemic Stroke Classification

The project titled “A Dual-Stream Deep Learning Architecture with Adaptive Random Vector Functional Link for Multi-Center Ischemic Stroke Classification” presents an intelligent and automated framework to improve the diagnosis and classification of ischemic stroke across diverse medical centers. Stroke, particularly ischemic stroke, remains one of the leading causes of mortality and long-term disability worldwide. Traditional classification approaches are often limited by dataset variability, noise, and the complex nature of brain imaging data. To address these challenges, this work integrates a dual-stream deep learning architecture with an Adaptive Random Vector Functional Link (ARVFL) network to enhance learning efficiency and robustness. The dual-stream network leverages convolutional layers to capture both local spatial details and global contextual features from medical images, while ARVFL acts as an adaptive classifier that improves generalization across heterogeneous datasets. The system is implemented using a Flask-based web application, enabling real-time stroke image uploads, classification into categories such as Bleeding, Ischemia, and Normal, and providing user-friendly accessibility. Additionally, the model’s multi-center adaptability ensures reliability when trained and tested on datasets from varied medical institutions, overcoming limitations of single-source training. With its scalable design and robust classification accuracy, the proposed system demonstrates potential to support neurologists in early detection, reduce misdiagnosis, and enhance patient care outcomes in clinical practice.

Keywords

Ischemic Stroke, Dual-Stream Deep Learning, Adaptive Random Vector Functional Link (ARVFL), Multi-Center Classification, Medical Image Analysis, Flask Application, Stroke Diagnosis.