This repository contains the implementation of a deep learning framework for modeling Huntington's Disease (HD) progression using point cloud representations of subcortical brain structures. We developed and trained a PointNet-based architecture to learn anatomical shape descriptors that capture subtle morphometric deformations associated with HD severity.
The core is a discriminative PointNet model trained to predict the Prognostic Index Normalized (PIN) score—a validated continuous measure of HD progression—from point cloud representations of segmented subcortical structures.
These learned shape descriptors were then integrated into a conditional generative model (cVAE) for forecasting clinical and volumetric biomarkers at follow-up.
This PointNet implementation was forked and adapted from the following repository:
We gratefully acknowledge the original authors for their open-source codebase, which served as the foundation for our customizations.