Published papers and studies still in progress, from my research at Keck School of Medicine of USC. Each paper links to its full write-up.
A multimodal vision-language model for Alzheimer's disease classification and question answering
A missing-modality-aware encoder over T1 MRI, DTI FA maps, and clinical scores that diagnoses from any available subset, with a retrieval-augmented language layer that surfaces comparable cases. Validated zero-shot on OASIS-3.
2,363
ADNI subjects
91.3%
CN vs Dem Bal. Acc.
78.7%
OASIS-3 zero-shot
Self-supervised 3D Vision Transformer encoders for brain MRI
Pre-trains 3D Vision Transformers on unlabeled brain MRI by predicting hidden regions in embedding space, then probes them for Alzheimer's staging, dementia detection and brain age. A variational objective beats the deterministic baseline on all three tasks at 25% of the pre-training data but loses at full data. Cross-cohort transfer is in progress.
16,024
T1 MRI scans
4
Cohorts
82
Pre-training runs