PD is one of the fastest-growing neurological diseases, with projections estimating 25.2 million people living with it by 2025 (a 112% increase from 2021) mainly due to the aging of the world’s population. Diagnosing PD at early stages is crucial to administer targeted treatment to patients and deep learning can aid PD diagnosis. However, SOTA models often fail to merge multiple neuroimaging modalities or to compensate for the absence of these in patient data. Furthermore, existing severity and progression prediction methods often do not take full advantage of data-rich neuroimaging modalities, favoring the analysis of peripheral symptoms such as speech, movement, and handwriting.