329-OR: Accuracy of Dietary Records (DR) Using a Smartphone App Is Greatly Improved by User’s Adjustments Compared with Artificial Intelligence (AI) Alone

Although DR are important in managing diabetes, the traditional method, such as weighing foods and calculating nutrients, is burdensome to both patients and medical care providers. Although photo DR with smartphones using AI to identify the dish and calculate nutrients can be expected to reduce this burden, accuracy not been clarified. Furthermore, it is not clear to what extent accuracy improves when users and dietitians adjust the name of the dish and the portion consumed compared with AI only. In this study, photo DR taken with the smartphone app were analyzed by AI only, then adjusted by the users followed by further adjustments by a dietitian. We compared the nutrients calculated by AI, users, and dietitians to those by the traditional weighing method. Analyzed were 180 participants (mean age 58.7 y, BMI 26.7 kg/m2, 94/86 women/men, 22 with diabetes, energy intake [EI] 1640 kcal/day) The EIs (kcal/day) were calculated as 1502, 1619, and 1674 for AI, users, and dietitians, respectively, those were lower than that with the weighing method Spearman's rank correlation coefficients compared to the weighing method were 0.48, 0.88, and 0.88 for AI, users and dietitians, respectively, showing a weak correlation for AI and a strong correlation for users and dietitians. The user's adjustments greatly improved accuracy without increasing the burden on the dietitians.