Presentation Information

[SS21-03]Differentiable simulation of neurodegenerative disease progression for brain image-based inverse modeling

*Yohei Kondo1 (1. Nagoya University (Japan))

Keywords:

Neurodegenerative disease

In neurodegenerative diseases such as Alzheimer’s and Parkinson’s, pathological lesions typically emerge in localized brain regions and gradually spread to neighboring areas. To elucidate the mechanisms underlying this progression and to enable predictive modeling, reaction-diffusion models based on partial differential equations have been investigated. However, a critical gap between simulation and reality still remains due to the lack of quantitative methods for determining spatial heterogeneity of parameters. To address this issue, we have developed an inverse method that leverages spatial lesion maps obtained from PET (positron emission tomography) imaging to estimate parameter heterogeneity of models. Our approach employs an optimization scheme based on error back-propagation through time, minimizing the discrepancy between simulated lesion distributions and observed imaging data. In the forward model, we incorporate anisotropic propagation along neural fibers by utilizing an average diffusion tensor image derived from the Human Connectome Project (HCP) database. Based on the analysis of quantitative PET data from approximately 300 individuals, we confirm that this anisotropy improves the predictive accuracy of lesion distributions compared with the isotropic counterpart. By applying this method to brain images from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, we have successfully inferred the spatial distribution of the prion-like amplification rate of abnormal tau species, which are otherwise inaccessible through direct observation.