Neuro Swin-Cloud: A Hybrid Swin Transformer Framework for Volumetric MRI Analysis

Abstract

Alzheimer’s disease is the most common form of dementia among older individuals and poses a major global health challenge, with nearly 10 million new cases reported annually. This progressive neurological disorder leads to gradual neurodegeneration, beginning with mild memory impairment and advancing to severe cognitive decline, reduced social interaction, and loss of environmental awareness. According to Alzheimer Disease International (ADI), approximately 75% of dementia cases remain undetected, making early diagnosis critically important yet challenging. Currently, effective diagnostic and treatment solutions for halting disease progression are limited. In this work, high-quality Magnetic Resonance Imaging (MRI) scans are processed using a pretrained Shifted Window (SWIN) Transformer to extract high-dimensional visual features that capture subtle neurostructural variations across different stages of Alzheimer’s disease, including Mild, Moderate, Normal, and Very Mild. These extracted features are evaluated using baseline classifiers such as Generalized Linear Model (GLM), Generalized Linear Vector Quantization (GLVQ), and Perceptron for comparative analysis. The proposed Optimal Rule Fit (ORF) classifier outperforms these methods by effectively modeling complex, non-linear relationships within the feature space. A Tkinter-based graphical user interface enables dataset upload, preprocessing, model training, evaluation, and visualization of confusion matrices and ROC curves. The system also supports remote prediction, allowing cross-device MRI-based diagnosis. Additionally, Explainable Artificial Intelligence techniques provide interpretable insights into brain atrophy, ventricular enlargement, and disease severity, enhancing clinical transparency and trust