Related Experiment Video
Updated: Aug 6, 2026
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
PatientSpace: A multimodal graph-based latent representation framework for modeling neurodegenerative disease
Dorian Manouvriez1, Grégory Kuchcinski2, Simon Lecerf3
1Univ. Lille, CNRS, Inserm, CHU Lille, Institut Pasteur de Lille, US 41 - UAR 2014 - PLBS, Lille, F-59000, France.
Neuroimage
|July 18, 2026
Summary
PatientSpace models neurodegenerative disease heterogeneity using multimodal neuroimaging. This interpretable framework links brain imaging to disease subtypes and predicts progression in Alzheimer's disease and frontotemporal dementia.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Neurology
Background:
- Neurodegenerative diseases like Alzheimer's (AD) and frontotemporal dementia (FTD) show significant heterogeneity.
- This complexity hinders accurate diagnosis, subtype classification, and prognosis prediction.
Purpose of the Study:
- To develop PatientSpace, a novel multimodal graph-based framework.
- To model and understand neurodegenerative disease heterogeneity using T1-weighted MRI and FDG-PET data.
Main Methods:
- PatientSpace utilizes a structured variational autoencoder to integrate multimodal neuroimaging features.
- It organizes patients in a latent space, constrained by clinical data and neuroimaging similarity.
- An interpretable patient graph is constructed, where proximity indicates biological similarity.
Main Results:
- PatientSpace identified distinct disease clusters in AD and FTD patients, correlating with neuroimaging patterns and clinical severity.
- Diagnostic classification performance matched state-of-the-art deep learning models.
- Graph-based inference accurately predicted structural volumes, metabolic activity, and cognitive severity.
Conclusions:
- PatientSpace offers an interpretable method to link multimodal neuroimaging to disease subtypes.
- The framework facilitates patient-level characterization and disease progression modeling in neurodegenerative disorders.
- It aids in predicting dementia conversion risk and longitudinal trajectories for mild cognitive impairment (MCI) patients.
