A Novel Approach to Calculating Energy Density from Food Images Reduces Analysis Time and Cost

Abstract

Traditional methods of self-reported food intake are characterized by limitations such as underreporting, high participant burden, and high cost. With the development of automated devices to capture food images and monitor food intake, an accurate and efficient method to estimate energy intake is needed. This study aimed to develop an accurate and time efficient method for estimating energy intake from food images by defining a simple and less burdensome way of estimating energy density (ED). Four experimental methods, exchange, food score-long, food score-short, and meal, were developed to estimate ED based on nutrient composition, water content, and relative proportion of foods in images, using different approaches. Three trained nutritionists analyzed 29 food images for ED using each method. All four experimental methods were compared to the full visual method in which a nutritionist estimated the portion size of each food consumed from dietary intake images and conducted data entry and analysis software. All experimental methods overestimated ED compared to the FVM but the meal method exhibited the closest agreement, lowest variance for ED, and significantly decreased analysis time by an average of 53 s/meal (p = 0.03). The meal method was used for full-scale validation by analyzing 213 food images against weighed food records. The meal method reduced analysis time by 69% (120 s; p ≤ 0.0001) and over-estimated ED by an average of 1.56 ± 3.17 J/g (p < 0.0001) compared to the FVM and 1.67 ± 3.09 J/g (p < 0.0001) compared to the WFR. The meal method is a novel and quick approach to calculate ED from dietary intake images.