Battling Diabetes through Food photography: An Image-based Utility Maximization Framework for Diet Diagnostics

Abstract

Diabetes is a leading source of health disorders, deaths, and healthcare costs worldwide, but can largely be prevented by leading a healthy lifestyle and diet routine. The rapid growth of Internet and mobile apps has made it possible for researchers to develop interventions that can collect fine-grained information on the diabetic patients’ eating habits to help them eat healthier. In this study, we propose the design and development of a novel framework for analyzing dietary data in the form of images and texts to predict health ratings of the food items, as well as recommend healthy food choices for diabetics in real time. We borrow from recent advances in deep learning and economic utility maximization models to automatically recognize different food components from the food photos, and subsequently make dietary recommendations that not only have high health utility but are also consistent with the individual’s taste preferences.