Summary: In a pilot study, a computer-vision food-photo app estimated energy intake about as well as the validated 24-hour recall method, but both underestimated true intake compared with objectively measured energy expenditure, so the app is a convenient alternative rather than a more accurate one.
PICO Summary
| Element | Detail |
|---|---|
| Population | Adults in the He Rourou Whai Painga dietary trial; pilot validation study, New Zealand. |
| Intervention | An iOS app using computer vision to estimate energy intake from food images, used for 7 days. |
| Comparison | Two 24-hour dietary recalls, and energy expenditure measured by indirect calorimetry and wrist accelerometry. |
| Outcome | Both the app and the recalls underestimated energy intake versus measured expenditure (mean bias -1814 kJ, p=0.005, for the app; -1715 kJ, p=0.029, for recall). The app and recall did not differ significantly from each other (mean bias 783 kJ; p=0.33), suggesting comparable performance. |
Expert Commentary
This is a sensibly designed validation pilot whose dual reference standard is its strength, since comparing an app not only against the 24-hour recall but also against objectively measured energy expenditure reveals what a method-versus-method comparison alone would hide. The honest and useful conclusion is one of parity rather than superiority: the image-based app matched the recall, but both underestimated true intake by roughly 1700 to 1800 kJ, which is the well-known underreporting that dogs all self-monitoring. The important interpretive point, which the post makes well, is that this shortfall reflects behaviour, forgetting to photograph meals or eating without recording, rather than a failure of the algorithm, so technology improves convenience without solving the fundamental capture problem. Limitations include a small motivated sample already enrolled in a diet trial, a single app, and a 7-day window. Can I use this with my patients? Yes, with realistic framing. For patients who will photograph meals, such an app is a lower-burden alternative to food diaries that may aid adherence in weight or diabetes management, but I would counsel that it, like all methods, underestimates intake by perhaps 15 to 20%, and is best used for tracking change over time rather than precise calorie counting.
References
Lee L, Bishop R, Stanley J. An automated image-based dietary assessment application: a pilot study. J Nutr Sci. 2025;14:e75. doi:10.1017/jns.2025.10045