A repository for comparative analyses across EuroPOND models.
- Model Staging using TADPOLE data
| Links | Public | Description | Input data type | |
|---|---|---|---|---|
| EBM | github | Yes | Event-based Model | Scalar biomarker values; cross-sectional |
| pyEBM | github | Yes | Toolbox for event-based models (EBM and dEBM) | Scalar biomarker values; cross-sectional |
| Deformetrica | gitlab website documentation |
Yes | Morphometry software. Works with image or mesh data. Can estimate anatomical registrations, regressions, cross-sectional & longitudinal atlases. | |
| LeasPy | gitlab | No | Learns the spatiotemporal patterns from longitudinal scalar measurements. | Scalar biomarker values; longitudinal |
| Clinica | gitlab website |
Yes | Clinica is a software platform for clinical research studies involving patients with neurological and psychiatric diseases and the acquistion of multimodal data (neuroimaging, clinical and cognitive evaluations, genetics...), most often with longitudinal follow-up. | |
| DEM | TBA | No | Differential Equation Model | Scalar biomarker values; longitudinal |
| SuStaIn | TBA | No | Subtype and Stage Inference | Scalar biomarker values; cross-sectional |
| DIVE | github (forked from Raz) |
Yes | Vertex Clustering Model | Images; longitudinal |
| Gaussian process regression model | github (forked from Marco) |
Yes | Gaussian Processes for temporal analysis of clinical data | Scalar biomarker values; cross-sectional or longitudinal |
Additional models of interest:
- Ageing trajectories:
- Leon's Multi-Task Learning trajectory model
- Eline's mixed (effects) models
- Off-the-shelf machine learning classifiers:
- SVM, MKL, etc.
- Point: for comparison of clinical classification (AUC)
- Stratified cross-validation. Sets defined by one person, then shared.
- Probably don't need to balance/stratify, due to TADPOLE numbers
- Evaluation measures:
- Appropriate TADPOLE metrics: AUC for clinical classification, too.
- (Regression) Correlations between model stages
- Hold-out longitudinal consistency:
- "Test" data: N (=200?) individuals having 2x2 timepoints (two disease staging opportunities) each
- Preferably not those having more than 4, so as not to reduce the data for unsupervised spatiotemporal models (Inserm)
- See python notebook for details