Automatic Estimation of the Nutritional Composition of Foods as Part of the GlucoseML Type 1 Diabetes Self-Management System

Abstract

The daily care of type 1 diabetes has been considerably improved through the increased adoption of continuous glucose monitoring, continuous subcutaneous insulin infusion, and precise behavioral monitoring (diet, physical activity) mHealth solutions. In this study, we present the food recognition and nutrient estimation components of the GlucoseML system; a type 1 diabetes self-management system relying on short-term predictive analytics of the glucose trajectory. A computer-vision-based approach is outlined combining image processing and machine learning to plate detection, food segmentation, food recognition and volume estimation of a plate's content. The systematic collection of an annotated Greek food images dataset allows the evaluation of the proposed methodology.