Learning Riemannian metric for disease progression modeling

Authors: Samuel Gruffaz, Pierre-Emmanuel Poulet, Etienne Maheux, Bruno Jedynak, Stanley DURRLEMAN

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The metric update allows us to improve the forecasting of imaging and clinical biomarkers in the Alzheimer s Disease Neuroimaging Initiative (ADNI) cohort. Our results compare favorably to the 56 methods benchmarked in the TADPOLE challenge.
Researcher Affiliation Academia Samuel Gruffaz Inria Paris Ecole normale supérieure Paris-Saclay samuel.gruffaz@ens-paris-saclay.fr Pierre-Emmanuel Poulet Paris Brain Institute Inria Paris pierre-emmanuel.poulet@inria.fr Etienne Maheux Paris Brain Institute Inria Paris etienne.maheux@icm-institute.org Bruno Jedynak Maseeh Professor of Mathematical Sciences Fariborz Maseeh Hall, room 464P Portland, OR 97201 bruno.jedynak@pdx.edu Stanley Durrleman Paris Brain Institute Inria Paris Inserm, CNRS, Sorbonne University stanley.durrleman@inria.fr
Pseudocode Yes Algorithm 1 Geodesics Bending (Alternating maximization algorithm)
Open Source Code No All code will be available on Github in the near future.
Open Datasets Yes Data used in preparation of this article were obtained from the Alzheimer s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). ... The TADPOLE training set is composed of data from the first three ADNI phases (ADNI 1, ADNI GO and ADNI 2).
Dataset Splits Yes For both experiments, models are trained with a 5-folds cross-validation.
Hardware Specification Yes All the methods are developed in Python by extending the open-source Leaspy library (https://leaspy.readthedocs.io) created for DCM models and run on a 2.80GHz CPU with 16 GB RAM.
Software Dependencies No The paper mentions 'Python', 'Leaspy library', and 'scikit-learn library', but does not provide specific version numbers for any of these software components.
Experiment Setup Yes We choose σnoise = 0.01, ggen = gφ with φ(x) := exp(0.07(x + sin(x)) 1) and g = ., . for the data generation and Nc(σ = 0.08) 90 for the metric estimation... we choose k to be the Gaussian kernel... n MCMC = 200 (n MCMC = 10000 at the very first step...).