Technical Specifications.
Verified parameters and operational benchmarks for this release.
Tech Stack.
Features Checklist.
Core capabilities, verified deliverables, and technical assets included with this release.
- 3D SFCN Alzheimer's classification
- 5-fold model ensemble
- Federated 3D SFCN
- Radiomics + Logistic Regression
- 17-feature radiomics analysis
- NIfTI MRI upload
- 3D MRI preprocessing
- MRI normalization and resizing
- Dementia probability prediction
- MRI slice visualization
- Multiple model selection
- Cross-fold prediction comparison
- Model performance metrics
- Streamlit web interface
- OASIS-1 dataset support
System Requirements.
- OS
- Linux / Windows / macOS
- CPU
- 4+ cores recommended
- GPU
- NVIDIA GPU recommended for 3D SFCN inference
- RAM
- 8GB minimum, 16GB recommended
- VRAM
- 6GB+ recommended
- Python
- 3.10+
- Browser
- Chrome / Firefox / Edge
- Storage
- 5GB+ for application and models
- Internet
- Required for initial dependency installation
- MRI Input
- NIfTI (.nii / .nii.gz)
About this Project.
SLAM — Alzheimer's Detection is a research-oriented MRI prediction platform containing three different machine-learning approaches for binary dementia classification.
The application accepts structural brain MRI volumes in NIfTI format and provides prediction workflows based on 3D SFCN models, federated 3D SFCN, and radiomics features combined with Logistic Regression.
Work 1A — 3D SFCN Cross-Fold
Uses five independently trained SFCN checkpoints for MRI classification. Users can either run the best-performing fold or execute all five models and average their predicted probabilities.
Work 1B — Federated 3D SFCN
Provides inference using a federated SFCN model trained across five virtual clients over multiple federated rounds. The application resizes MRI volumes to 96×96×96 before inference.
Work 2 — Radiomics + Logistic Regression
Provides a traditional machine-learning workflow using selected radiomics features extracted from Brainstem, Left Hippocampus, and Right Hippocampus regions. The trained Logistic Regression pipeline produces a dementia probability and classification.
Visualization
The application automatically generates normalized MRI volume previews showing sagittal, coronal, and axial views.
The Logistic Regression workflow also provides a visualization of model coefficients for the selected radiomics features.
Intended Use
Machine-learning research Medical-imaging research MRI classification experiments Federated-learning demonstrations Deep-learning experimentation Radiomics research University research projects AI/ML education Research prototype demonstrations OASIS-1 experimentation
Included
Streamlit application 3D SFCN architecture Cross-fold inference workflow Federated SFCN inference workflow Radiomics + Logistic Regression workflow MRI preprocessing utilities NIfTI loading MRI visualization Model loading utilities Prediction interface Threshold configuration Probability visualization Model metrics interface Radiomics feature interface
Important Disclaimer
This project is a research and educational prototype and is not a medical diagnostic device.
Predictions should not be used to diagnose Alzheimer's disease, dementia, or any other medical condition.
The models are trained/evaluated using research MRI data and require independent clinical validation before any consideration of real-world clinical use.
Performance on OASIS-1 does not establish generalization to other hospitals, scanners, populations, acquisition protocols, or clinical settings. OASIS-1 is a research neuroimaging dataset containing structural MRI data across normal aging and Alzheimer's disease research cohorts.
The application is intended for experimentation, education, and research.