NeuroVascX: A Hybrid Transformer–Extreme Gradient Boosting Framework for Vascular Dementia Identification

Vascular dementia (VaD), a prevalent neurodegenerative disorder caused by cerebrovascular abnormalities, is the second most common form of dementia after Alzheimer’s disease. VaD requires early and accurate diagnosis to plan therapeutic measures as well as future care. The conventional methods of diagnosing which depend on the clinical examination and radiological evaluation are usually limited in sensitivity and prone to subjectivity. In order to address these limitations, we introduce a new hybrid deep learning architecture, NeuroVascX, which combines the use of a fine-tuned Swin Transformer, performing feature extraction on MRI scans, with the Extreme Gradient Boosting (XGBoost) classifier to predict accurate predictions. The system is trained and tested on publicly available vascular dementia dataset of Kaggle that will include structural MRI images which will be sorted in different classes Normal and Vascular Dementia. The preprocessing pipeline contains image normalization, denoising with Non-Local Mean filtering, as well as augmentation to add generalization to the model. Transformer (Swin) encodes rich contextual features in explicitly normalized slices in MRI and XGBoost uses such embedding to classify. The model attained an accuracy of 89.87%, F1-score of 91.51% and Matthews correlation coefficient (MCC) of 0.89, and out-performed numerous benchmark models such as ResNet50, VGG16, and standalone XGBoost. Moreover, the inference time value of 14.78 milliseconds shows that it is possible to use this application in a real-time clinical setting. The suggested NeuroVascX framework offers a potentially viable AI-supported technique that can enhance the accuracy and speed of screening vascular dementia and is interpretable, potentially making routine neuroimaging utilization a possibility in real-life telemedicine platforms.