Key points are not available for this paper at this time.
Objectives: Recent evidence indicates that Central Asia (CA) experiences one of the highest burdens of diet-related deaths globally. The lack of a representative food dataset and current labor-intensive dietary assessment methods underscore the need for a contemporary and efficient approach to dietary assessment. Artificial Intelligence (AI) facilitates accurate and rapid dietary assessment through computer vision and smartphone technology for food image recognition. Several datasets have been compiled, with a focus on different cuisines. However, a dataset specific to Central Asian foods has been lacking to date. To address this gap, our research team has embarked on an inaugural attempt to construct the Central Asian Food Dataset (CAFD) encompassing 42 distinct food categories, comprising over 16,000 images showcasing unique national dishes from the region 1. Building on this dataset, this project now aims to create a digital food atlas of commonly consumed meals in CA in order to improve the accuracy of visual food portion size estimation, which is crucial for understanding dietary intake in research and public health contexts. Methods: A list of 120 commonly consumed foods and their corresponding portion sizes was derived from a priori from existing knowledge and literature. These food items will then be weighed and presented in three different portion sizes before being photographed by an Intel Realsense D455 RGB-depth camera, under identical conditions. A validation study will be conducted to evaluate the portion sizes by comparing them with a series of photographs for each food item. We will also use a pre-trained model to estimate the portion size from food images. Results: Mean differences and standard deviations between estimated and actual portion sizes will be reported and these can provide insights into the reliability and variability of portion size perception. Conclusions: The validation of the methodology ensures the credibility and reproducibility of the digital atlas for future use. In summary, once further validated, this digital food atlas could be utilized to better estimate the nutritional content of a dish from visual data. Funding Sources: NU Faculty Development Competitive Research Program.
Chan et al. (Sat,) studied this question.