Synthetic datasets and their predictive models are all set for increased global applications in medicine. This surge is because of the tightening of privacy regulations and gains in model performance. Synthetic datasets based on a single cohort are disadvantaged due to sample-selection bias and hence its models fail to generalize especially in neuroscience. This calls for a novel framework to integrate multiple cohorts because it can address both the single cohort's limitations and enable precision medicine when neurodegenerative datasets are employed. I propose to develop neuro-data-specific methods forming a framework through this project (NeuroSynth). I hypothesize that these novel methods can significantly improve the state-of-the-art by overcoming the challenges specific to 1) fusing data from different modalities, 2) combining data from different research centers, and 3) tracking clinically relevant parameters longitudinally.

2022

2025