Shape matters: Predicting Huntington's disease using progression modelling

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Biomedical Data Science

Background:

• Current models for Huntington's Disease (HD) progression in clinical trials do not utilize detailed striatal morphometry (shape information).
• This limits the precision of participant selection and treatment outcome assessment in clinical research for persons with HD (PwHD).

Purpose of the Study:

• To investigate the utility of subcortical brain shape descriptors in modeling HD progression.
• To assess if shape information can improve predictive models for disease biomarkers compared to volumetric data.

Main Methods:

• Validated a deep neural network to extract shape descriptors from subcortical structures in 2,932 brain scans from 615 PwHD across three longitudinal datasets.
• Trained a conditional generative model using shape descriptors, volumetric, genetic, and clinical data to predict disease progression biomarkers.

Main Results:

• Anatomical shapes of key subcortical structures (putamen, lateral ventricle, pallidum, caudate, thalamus, accumbens) strongly correlated with HD progression.
• Aggregated shape descriptors via principal component analysis showed higher correlation with disease stage (ρ = 0.72) than volumetric measurements (ρ = 0.45).
• Incorporating subcortical shape into the generative model significantly improved predictive performance over models using only brain volumes.

Conclusions:

• Subcortical brain shape is a significant correlate of HD progression and captures finer within-stage variability.
• Shape-based models enhance the predictability of HD biomarkers, offering potential for more precise clinical trial participant selection.
• This approach could lead to more objective post-intervention assessments of treatment efficacy in future HD clinical trials.