Matthew Salvino: Applying Machine Learning to Health Care
A profile of an undergraduate engineering research fellow at Duke University.
- Major: Biomedical Engineering
- Advisor: Lawrence Carin, James L. Meriam Distinguished Professor of Electrical and Computer Engineering and Duke’s Vice President for Research
- Project: Deep Learning Image Analysis for Pathology
What drew you to Duke Engineering?
I always knew that I wanted to go to medical school. So, one of the biggest questions for me was what path would I want to take to get to medical school? I have always been technically minded, and I was drawn to engineering. As I did more research into Duke, I came to understand that it had a great biomedical engineering program, and one of the things that have always fascinated me is the intersection of medicine and technology.
Who is your Pratt Research Fellows mentor and what kind of work do you do?
The research I’m doing for the Pratt Research Fellows program is in the lab of Larry Carin, who is a professor of computer and electrical engineering as well as vice president for research at Duke. The work pertains to applying machine learning to health. I’ve worked on a number of different projects; one of the first projects I worked on was developing an automated method of detecting diabetic retinopathy. The program takes images of the retina and, using convolutional neural networks, can assess not only the presence of diabetic retinopathy but also its extent. That can be used as a very important prognostic tool.
In another project, we’re looking at fine-needle aspirations of thyroid cancer. These pathology slides have hundreds of millions of pixels. It is very time-consuming to analyze due to the size of each slide. One of the things we’re focusing on is taking these images at lower resolutions, and separating Bethesda 6 slides, one of the most malignant cases of thyroid pathology, from non-Bethesda 6 slides.
We’re also developing a pathologist tool for prostate cancer classification. The goal is to determine whether pathologists can accurately classify prostate cancer using only areas of interest from a prostate cancer biopsy. We developed a tool in which a classifier extracts different regions of interest from a biopsy. We identified the regions that have the highest predictions of being malignant, and we’ll only show those regions to the pathologist. The goal of this project is to determine whether this is a feasible method—whether a pathologist can accurately predict or classify prostate cancer using only certain patches from a particular biopsy, with the ultimate goal of saving time in classifying these cancers.
What is one thing that you have enjoyed most about your research?
The cool thing with this research is that you get to see the clinical applications. I’ve interacted with pathologists and different people at Duke Health to understand the need for this technology and how they might incorporate it into their daily practice.
How were you matched with Dr. Carin?
I knew who I wanted to work with and what project I wanted to work on before I even applied to the program. So, I applied with that knowledge in mind. I know some other people have a general idea of the type of project they want to work on or the professor they want to work with. The coordinators and organizers of the program do a great job of matching people to interesting and impactful projects.
Would you recommend the Pratt Research Fellows program?
Absolutely. I think it’s a great opportunity to explore a particular field that you wouldn’t necessarily get exposed to through your major. For me, as a biomedical engineer, I knew I wouldn’t necessarily get a foundation in machine learning, so I used it as an opportunity and a learning experience, to get exposure to this field and to see its applications in medicine. In addition, the class credit and the summer funding from the program really give you the tools, resources, and focus to put meaning into the work you are doing.