1305-GRP Adaptive Glucose-Peak Prediction for Diabetes Using Meal Photos and Continuous Glucose Monitoring

Presenter Information

Location

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Document Type

Event

Start Date

20-11-2026 2:00 PM

Description

Postmeal glucose responses vary with meal composition and personal glucose dynamics. This project combines continuous glucose monitoring (CGM) with nutrition estimated from meal photographs to predict the highest glucose within a two-hour eating-occasion window. Data were collected in collaboration with Emory University. The model uses current glucose, recent trends, historical response patterns, and meal nutrition, with predictions updated at meal time and after 15, 30, and 45 minutes. A retrospective participant-held-out evaluation included 62 people with diabetes and 717 meal-time eating occasions. Each participant’s population model was trained on other participants. Mean absolute error was 25.24 mg/dL at meal time, decreasing to 23.42, 18.91, and 14.67 mg/dL at the respective updates. Later estimates incorporate a larger observed portion of the same outcome window. These findings support further investigation of adaptive glucose-peak estimates for personalized nutrition feedback. Prospective evaluation is needed to establish clinical usefulness and address uncertainty in nutrition estimates and input timing.

This document is currently not available here.

Share

COinS
 
Nov 20th, 2:00 PM

1305-GRP Adaptive Glucose-Peak Prediction for Diabetes Using Meal Photos and Continuous Glucose Monitoring

https://campus.kennesaw.edu/colleges-departments/ccse/events/computing-showcase/fa26-cday-program.php

Postmeal glucose responses vary with meal composition and personal glucose dynamics. This project combines continuous glucose monitoring (CGM) with nutrition estimated from meal photographs to predict the highest glucose within a two-hour eating-occasion window. Data were collected in collaboration with Emory University. The model uses current glucose, recent trends, historical response patterns, and meal nutrition, with predictions updated at meal time and after 15, 30, and 45 minutes. A retrospective participant-held-out evaluation included 62 people with diabetes and 717 meal-time eating occasions. Each participant’s population model was trained on other participants. Mean absolute error was 25.24 mg/dL at meal time, decreasing to 23.42, 18.91, and 14.67 mg/dL at the respective updates. Later estimates incorporate a larger observed portion of the same outcome window. These findings support further investigation of adaptive glucose-peak estimates for personalized nutrition feedback. Prospective evaluation is needed to establish clinical usefulness and address uncertainty in nutrition estimates and input timing.