Mingui Sun, PhD, received a BS degree in instrumental and industrial automation in 1982 from the Shenyang Chemical Engineering Institute in Shenyang, China, and an MS degree in electrical engineering in 1986 from the University of Pittsburgh, where he also earned a PhD degree in electrical engineering in 1989. He was later appointed to the faculty in the Department of Neurological Surgery.
Dr. Sun’s research interests include neurophysiological signals and systems, biosensor designs, brain-computer interface, bioelectronics, machine learning and artificial intelligence. He has more than 460 publications.
Dr. Sun's publications can be reviewed through the National Library of Medicine's publication database.
Specialized Areas of Interest
Professional Organization Membership
Education & Training
- BS, Instrumentation/Industrial Automation, Shenyang Chemical Institute, 1982
- MS, Electrical Engineering, University of Pittsburgh, 1986
- PhD, Electrical Engineering, University of Pittsburgh, 1989
Research Activities
Automatic Carbohydrate Counting Using AI and Large Language Models
Accurate carbohydrate counting (CC) is one of the core dietary management skills for effective glycemic control in individuals with diabetes. However, the traditional approach largely relies on the patients’ estimation of food portion sizes and carbohydrate amounts. Thus, this approach is subjective and burdensome.
Recently, an image-based approach has been developed in which patients take pictures of their food using a smartphone or a wearable device. The resulting images are processed to identify food items, their portion sizes, and the amounts of carbohydrates consumed. Although the image approach is more objective, it is still burdensome due to the required manual procedures to handle image data and search a food database.
More recently, AI technologies, especially the Large Language Models (LLMs), emerged which shined new lights on CC. LLMs can automatically recognize foods from images, estimate their portions, and then produce CC results without requiring a database lookup. Although the AI approach holds a strong promise for CC, the accuracy of the current LLM results remains unsatisfactory.
Dr. Sun is currently investigating two primary causes of errors, one is a questionable food shape fitting method utilized by LLMs to estimate food portion size, and the other is an over-reliance of LLMs on common object dimensions obtained from training data. Dr. Sun believes that the accuracy of the LLM-based CC system can be improved substantially by mathematical modeling the food shape and supplying case-specific information to LLMs through appropriately designed prompts.