Fuzzy-Augmented MobileNetV2 for Four-Stage Alzheimer's Disease Classification from MRI: A Reproducible Pipeline with Honest Benchmarking

AlzDetect is a lightweight and reproducible pipeline for four-stage Alzheimer's disease classification from MRI scans. The system combines a frozen MobileNetV2 feature extractor, feature-space SMOTE balancing, a Mamdani fuzzy resampling controller, and a trainable Takagi-Sugeno-Kang fuzzy inference layer. The study reports controlled ablation experiments, deployment details, reproducibility resources, and a discussion of domain-shift limitations. Code, trained weights, and documentation are publicly available through the associated GitHub repository.