SOE Engineer Magazine_Sept 2022 update


A Conversation about Advancement of AI in Medical Imaging Analysis


AI is changing the landscape in almost every field. During the 2021 IEEE International Conference on Bioinformatics and Biomedicine, Dr. Chih Lai (right), professor of Graduate Programs in Software, and his colleagues presented their recent research on the effects of AI on fish embryo development and its implications for medical imaging analysis. St. Thomas Engineer spoke with Lai, and a research assistant on the project, Akhil Ambekar ’20 (left) MS in data science. Ambekar is now an AI health fellow at Duke University.

How are you using data science to further research knowledge in medical image analysis? Chih Lai: My AI research team, together with scientists from Helmholtz Center for Environmental Research (UFZ) in Germany and University of Barcelona in Spain, used AI to study the toxicity impact to embryonic development due to pharmacological treatment. With AI, we were able to analyze a larger number of scenarios (a total of 4,080 scenarios – eight different organs, 15 different chemicals, two different exposure times and 17 concentration levels) and increased the image precision, showing subtle morphological changes.*

What is the future of AI in medical imaging analysis? Lai: Increasingly, toxicity/ pharmaceutical research is

Akhil Ambekar: The imaging techniques and the deep learning methods learned while working on the fish embryo and toxicity project played a major role in my transition from a research assistant at University of St. Thomas to an AI health fellow at Duke University. What I learned was extremely relevant to what is currently being used in the real world. Now at Duke, I am currently working on several exciting projects made possible due to artificial intelligence. One project is focused on automating the process of identifying scarred regions in the kidneys. There is both diagnostic and prognostic relevance to identifying these scarred regions, but without a deep learning model, this can be time-consuming, and have limited accuracy and reproducibility of observations.

combining traditional data analysis with AI technology because AI can recognize and detect the areas more likely to lead to fruitful results without brute force trial and error. While AI is not a replacement for human experts, technology is advancing medical imaging analysis, helping accelerate our understanding of toxic impacts to fish embryos. * Parts of the system developed by the team in automating this analysis process have been published and released to the toxicology/biology community under the GNU license.

Page 18

Powered by