(OR07-02-26) Photograph-Based AI Features in Calorie-Tracking Apps Underestimate Energy Content of Meals

Saturday, July 25, 2026

4:42 PM - 4:54 PM ET

Location: National Harbor 4-5

Presenting Author(s)

OC

Olivia Charles, BS

Research Fellow
National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health
Bethesda, Maryland, United States

Disclosure(s): No relevant financial relationship(s) with ineligible companies to disclose.

Co-Author(s)

EF

Eleanor U. Flacke, BS

Postbaccalaureate Intramural Research Fellow
National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health
Bethesda, Maryland, United States
ST

Sara Turner

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
SY

Shanna Yang

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
KA

Kayla Airaghi

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
NV

Natalie Vallone

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
LH

Lindsay Herra

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
Valerie L. Darcey, PhD, MS, RD, LD (she/her/hers)

Stadtman Investigator
National Institutes of Health, NIDDK
Bethesda, Maryland, United States
SC

Stephanie T. Chung

National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health, United States
Aaron Hengist, PhD (he/him/his)

Postdoctoral Visiting Fellow
National Institutes of Health
Bethesda, Maryland, United States

Disclosure(s):

Olivia Charles, BS: No relevant financial relationship(s) with ineligible companies to disclose.

Valerie L. Darcey, PhD, MS, RD, LD: No relevant financial relationship(s) with ineligible companies to disclose.

Aaron Hengist, PhD: No relevant financial relationship(s) with ineligible companies to disclose.

Objectives: We examined accuracy of photograph-based AI features from 4 calorie-tracking apps to estimate meal nutritional composition. We hypothesized that photograph-based AI features would underestimate energy and macronutrient content of meals.

Methods: Nutrient-controlled meals were planned using national food composition tables for a controlled feeding study. Photographs of 102 prepared meals were analyzed using commercially available apps: A) MyFitnessPal, B) LoseIt!, C) CalAI and D) Appediet. Differences in energy and macronutrient content between apps and food composition tables were compared with paired t-tests. Agreement and bias were measured using limits of agreement (LoA) and simple linear regression from Bland-Altman analyses. Consistency was assessed by Intraclass Correlation Coefficients (ICC).

Results: Mean (range) energy and macronutrient content of meals was 918 (592-1211) kcal, 60 (4-189) g carbohydrates, 61 (21-108) g fat, and 36 (25-51) g protein. All apps underestimated mean (95% CI) energy content: A -327 (-385 to -269), B -333 (-383 to -282), C -345 (-392 to -296), and D -252 (-295 to -210) kcal (all P < 0.001). LoA for apps A-D were -884 to 246, -835 to 170, -821 to 132, -672 to 167 kcal. Apps A (R²=0.25, p< 0.001) and B (R²=0.05, p< 0.025) displayed significant bias with higher energy meals displaying less error than lower energy meals. Apps B and C underestimated carbohydrates by ~14 g (both P < 0.001), without significant differences for A or D (P≥0.07). All apps underestimated fat by ~30g (P<0.001). ICC for carbohydrates were moderate-to-good across apps (all ≥0.69), but consistency was poor (all < 0.40) for other outcomes

Conclusions: Photograph-based AI features in apps underestimated energy content of meals by ~33%, similar to previously reported underestimation using self-report diet assessment methods. MyFitnessPal and LoseIt! estimated energy content of higher energy meals more accurately than lower energy meals, and apps estimated carbohydrates more consistently than other macronutrients. Future work should assess the utility of photograph-based AI to decrease burden of traditional diet assessment methods.

Funding Sources: This research was supported by the Intramural Research Program of the National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) within the National Institutes of Health (NIH).