Alzheimer's Disease Detection System
A deep-learning classification system developed as academic research, using convolutional neural networks over MRI imaging to distinguish between stages of cognitive decline. Built as a study in medical image classification — not as a clinical or diagnostic product.
Project detail
- Type
- Academic Project
- Category
- Machine Learning
- Year
- 2024
- Technology
- PythonTensorFlowKerasOpenCVNumPyscikit-learnMatplotlib
Distinguishing early stages of Alzheimer's from MRI imaging is a subtle, high-skill classification task. Public research datasets are also heavily imbalanced — mild and very mild cases are far scarcer than non-demented ones — which makes naive models look accurate while failing on precisely the cases that matter most.
The pipeline standardises MRI slices through normalisation and augmentation, then trains a convolutional architecture using transfer learning from pretrained vision backbones. Class imbalance is addressed through weighted loss and targeted augmentation of minority classes. Evaluation deliberately avoids headline accuracy in favour of per-class precision and recall and a confusion matrix, and Grad-CAM overlays are generated so the regions driving each prediction can be inspected rather than trusted blindly.
What the system does
MRI preprocessing pipeline
Normalisation, resizing and augmentation applied consistently across training and evaluation.
Transfer-learned CNN classifier
Pretrained vision backbones fine-tuned for multi-class staging of cognitive decline.
Class imbalance handling
Weighted loss and minority-class augmentation, so rare stages are not sacrificed to headline accuracy.
Explainability
Grad-CAM heatmaps show which regions of a scan drove each prediction.
Honest evaluation
Per-class precision, recall and confusion matrices reported instead of a single accuracy figure.
What it demonstrates
Stated as capability rather than invented metrics — we only publish numbers we can stand behind.
Scope
Academic research project — not a clinical diagnostic tool
Method
Transfer learning with explicit class-imbalance handling
Transparency
Grad-CAM explainability over every prediction
Other case studies
Have a project like this in mind?
Book a free consultation and we'll walk through the approach, the scope and what it would realistically take.
Prefer to just message? Chat with us on WhatsApp.