manuelmoracartagena/alzheimer-mri
Reproducible Deep Learning pipeline for Alzheimer's stage classification using CNN, ResNet, ViT & HybridRViT.
What's novel
Reproducible Deep Learning pipeline for Alzheimer's stage classification using CNN, ResNet, ViT & HybridRViT.
Code Analysis
13 files read · 4 roundsA PyTorch-based deep learning pipeline that classifies Alzheimer's Disease stages from brain MRI scans using multiple architectures (CNNs, ResNet, ViT, HybridRViT) with configurable training, cross-validation, and WandB experiment tracking.
Strengths
Clean config-driven architecture with a well-organized model factory pattern, and genuinely implemented (not just imported) neural network architectures including a novel HybridRViT that feeds ResNet feature maps directly into a ViT encoder. The code is well-documented with type hints and docstrings throughout.
Weaknesses
Zero test coverage, no learning rate scheduler, and the dataloader's error handling silently returns a black image with label -1 which would corrupt training. The README claims 'early stopping' but the code simply runs all epochs and selects the best model by validation loss.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
25
Contributors
1
Files
37
Active weeks
5
Repository
Language
Python
Stars
1
Forks
0
License
MIT