Feed

Accurate assessment of an individual's diet is vital to study the effect of diet on health. Image-based methods, which use images as input, may improve the reliability of dietary assessment. We developed an iOS application that uses computer vision to identify food from images. This study aimed to assess the accuracy of energy intake (EI_app) estimates from the application by comparing them to estimated energy expenditure (EE) and to the EI estimates from a validated dietary assessment tool, the 24-h recall (EI_recall). Participants were recruited from a randomised controlled trial called He Rourou Whai Painga. Participants recorded all intake over 7 d using the application, which provided a mean daily EI; this was compared to the EI estimated by two 24-h recalls. The EI from the application and the recalls were compared to EE, estimated using indirect calorimetry and wrist-worn accelerometry. EI estimates from the application and the 24-h recalls were lower than EE, with a mean bias of -1814 kJ (95% CI -3012 to -615, p = 0.005) and -1715 kJ (95% CI -3237 to -193, p = 0.029), respectively. The mean bias between EI from the application and the 24-h recall was 783 kJ (95% CI -875 to 2441, p = 0.33). This suggests that the EI estimates from the application are comparable to the 24-h recall method, a validated and widely used tool in nutritional research.