Imagecarb: An Intelligent Application for Carbohydrate Estimation in Type 1 Diabetes Management

Abstract

Type 1 Diabetes Mellitus (T1DM) requires strict carbohydrate counting to adjust insulin doses and maintain glycemic control. Manual carbohydrate estimation is time-consuming and prone to human error, creating a need for automated tools. This study presents the development of ImageCarb, an intelligent application implemented in MATLAB that automatically segments food items from digital images and estimates carbohydrate content. Images were standardized with a white round plate, single food type per plate and light background to ensure reliability. Two segmentation approaches were implemented: A manual Region of Interest (ROI) method and a fully automated segmentation method using colour space conversion (RGB to HSV and LAB), adaptive thresholding and morphological operations. Nutritional data were integrated from standardized databases photographic manual for food quantification [1]. Tests were performed with representative Portuguese foods, including carrots, courgettes, red cabbage, peas, beans, ham and cold cuts. Results demonstrated that the automated method achieved a discrepancy of only 2-3% compared with the manual method, validating its accuracy and robustness. Carbohydrate estimates were consistent with reference nutritional tables and the system successfully provided macronutrient breakdown (carbohydrates, proteins, lipids). The ImageCarb application thus offers a user-friendly tool to support carbohydrate counting in T1DM patients, with potential to improve glycemia management and quality of life. Future directions include expanding the system to complex mixed meals, integrating real-time continuous glucose monitoring and implementing a mobile application interface.

Keywords: Type 1 diabetes; Carbohydrate counting; Digital image processing; Food recognition

Abbreviations: ADA: American Diabetes Association; CHO: Carbohydrates; GUI: Graphical User Interface; HSV: Hue Saturation Value; LAB: Lightness, A (greenness-redness) and B (blueness-yellowness); T1DM: Type 1 Diabetes Mellitus; RGB: R (Red), G (Green) and B (Blue), ROI: Region of Interest