Research Assistant

Working as an RA as part of my PhD. I have been funded by the NIH, primarily focusing on clinical AI in data-scarce domains.

Extracorporeal Membrane Oxygenation (ECMO)

This is the primary domain I have been looking at. I have been working with clinicians at UTSW to model the risk of neurological injury in patients on ECMO using high-frequency physiological and laboratory temporal data. The domain is data scarce and high-dimensional, making typical machine learning and deep learning methods not work in the domain. We have been looking at LLMs as approximate knowledge sources to generate an initial hypothesis to solve the problem.

Adverse Pregnancy Outcomes (APO)

I have been working on the APO datasets for the nuMoM2b study and evaluating the LLM’s ability for causal modelling using the domain. While the domain is not as data-scarce as ECMO, it is a very unbalanced domain with the APOs, like gestational diabetes, present in a small percentage of the population

Life-Space Assessment

We have been working with researchers from the University of Minnesota to model life-space scores for rural and Indigenous populations in the US and Canada. Lifespace Assessment is mainly focused on urban population centers, leading to a lack of data for our population of interest, making it a challenging problem to tackle/

Alzheimer’s Verbal Fluency Tests

We have also been looking at Audio Transcripts for verbal fluency tests for Alzheimer’s disease (AD) to model the relationship between mild cognitive impairment (MCI) and AD markers.

Appendectomies

We are working with appendectomy operation notes to see if we can induce a general plan for the procedure from these notes using Large Language Models