Please provide a brief overview of your research and/or creative work. You are welcome to focus on your most current work or, if you prefer, any particular work you’d like to highlight.
Broadly speaking, my research focuses on the development of interpretable statistical and computational techniques to analyze imaging and complex network data. My current work generally falls into three themes:
- Modeling and Analysis of Brain Imaging Data
- Modeling Social Dynamics and Influencers on Social Media
- Interpretable Machine Learning
I could spend a lot of time talking about my work in each of these areas, but I want to highlight two of my recent publications that have really had an impact on my current direction of work:
- Torbati, M.E., Minhas, D.S., Laymon, C.M., Maillard, P., Wilson, J.D., Chen, C- L., Crainiceanu, C.M. DeCarli, C.S., Hwang, S.J. and Tudorascu, D. (2023) MISPEL: A deep learning approach for harmonizing multi-scanner matched neuroimaging data. Medical Imaging Analysis 89, 102926
- Wilson, J.D., Gerlach, A., Aizenstein, H., and Andreescu, C. (2024) Sex matters: Acute functional connectivity changes as markers of remission in late-life depression differ by sex. Molecular Psychiatry 28 (12), 5228-5236.
These papers, different as they may seem, actually both highlight an important and often overlooked challenge in data science and analytics, which for lack of a better phrase I’ll call the “wrong data problem.” Particularly in today’s era of accessible deep learning apps like DeepSeek and ChatGPT, one could reasonably believe “the more data the better” for building predictive models (e.g., OpenAI’s strategy for ChatGPT). Though it is useful to have more information about an outcome we would like to predict, it is wrong to think that all available data should be used in the prediction, or that adding new data will necessarily help.
What I have learned from the above research is that we must understand and compensate for possible differences in data obtained from disparate sources, technologies, or people. The first publication above, for example, focused on the differences in brain images obtained from scanners from different companies (e.g., Philips, Siemens, etc.). The idea in that paper was to develop deep learning strategies to quantify and account for technical variability of images so that data from disparate scanners could be effectively analyzed together. The second publication I mentioned assessed sex-based differences in the brain’s response to anti-depressant medication for patients with late-life depression. We showed that the 24- hour change in how a person’s brain reacts to a new anti-depressant medication (a) was strongly predictive of whether or not that person would be remitted to the hospital in need of different medication, and (b) that the predictive ability of the brain scans was highly dependent on the person’s biological sex. In fact, remission prediction was 20% worse when males and females were put into the same model rather than in separate models. This work highlights the fact that in understanding the brain’s response to antidepressant treatment, medical professionals must account for sex differences; furthermore, not doing so may lead to devastating outcomes due to depression going improperly treated.
What inspired you to pursue this area of study or creation?
I have been interested in brain imaging analysis since I was a second year Ph.D. student at UNC Chapel Hill. I remember my first time diving into an imaging data set where the goal was to identify individuals with varying subtypes of ADHD (the “ADHD 200 challenge”). I threw every statistical method I knew at the time at this data and could do no better than a coin flip in making predictions. Though my predictions at the time were horrible, this experience drove me to want to figure out how to use more innovative methods in machine learning and network analysis, which I was studying in my Ph.D. thesis work, to make progress in the area. Since then, I have sought out and worked with many wonderful collaborators in psychiatry, neuroscience, biology, imaging, and physics from whom I have learned a tremendous amount about imaging analyses, neurodegenerative diseases, and mental illness. These collaborations have opened up incredibly exciting areas in medical research that I am excited to pursue.
What impact do you hope your work will have on your field and/or the broader community?
I am fortunate that I am a statistics and data science researcher and educator, where I have the opportunity both to work with incredible collaborators and to teach aspiring data scientists my own lessons learned in research. My overall hope is to show the broader community how to think about data and analyses, rather than just simply running the most accessible app or program to do the work for you. As I pointed out in my highlighted research, we cannot always assume that simply analyzing all the data we have our hands on is a good idea. Understanding individual differences, technological differences, and other sources of variation in the data we learn from is paramount to making informed decisions.
How has your involvement with CRASE influenced and enhanced your professional journey?
As a faculty member, it is easy to get carried away with preparing the next (hopefully) amazing lecture. Our devotion to excellent teaching and service to the university and broader community can make it challenging to make the time for research and scholarly work. Even when we do finally get that chance to sit down and work, it is easy to miss what amazing scholarly work everyone else at the university is up to. CRASE is a wonderful initiative at USF that focuses on motivating, making space for, and celebrating the research and scholarly work of faculty members at the university. In being part of CRASE, I have been able to witness the amazing research of my colleagues through meetings, writing workshops, and celebration events. It has opened my eyes to exciting interdisciplinary research opportunities at the university and has inspired me to investigate new areas in my own research that I wouldn’t have otherwise thought to do. I am excited to contribute to the CRASE initiative and hope to meet more of you through some of our events in the near future.