New research from the University of Sydney has found that developing nutrition apps that track food intake and manage weight requires improved artificial intelligence (AI) training.
The researchers initially screened 800 apps and selected 18 for further evaluation. These 18 apps, which included both AI-integrated and manual food-logging nutrition apps, were assessed for their ability to recognise ingredients and estimate their energy content.
The findings have been published in the journal Nutrients.
Lead author of the study, Dr Juliana Chen, a registered dietitian, lecturer and researcher in nutrition at the University of Sydney, said AI-integrated apps are more convenient than manual food logging, but suggested they should be used with caution.
“When patients or the public use apps to record their food intake or manage their weight, the process can often feel cumbersome,” Dr Chen said. “Adding AI features such as food image recognition has the potential to make the process much easier for everyone.”
“However, it's always important to double-check that the portions detected match what you actually ate. Some apps only identify foods, but others also estimate portion sizes and energy intake. So, if you're trying to lose weight, it's important to check that the app's estimates match what you actually ate.”
A key part of the study was to see how accurate and adaptable these apps were across three different eating plans: Western, Asian and Recommended (based on the Australian Dietary Guidelines) to ensure different cultural food preferences were taken into account.
Working under Dr Chen's supervision, nutrition master's students Xin-Yi Li, Annabelle Yin and Ha-Young Chey found that manual food tracking apps overestimated energy intake for Western diets by an average of 1,040 kilojoules, and underestimated energy intake for Asian diets and recommended diets by an average of 1,520 kilojoules and 944 kilojoules, respectively.
In contrast, the AI-integrated dining app often had difficulty accurately identifying the energy content of the Asian food mix, overestimating calories in beef pho by as much as 49% and underestimating calories in pearl milk tea by up to 76%, for example.
“AI-integrated nutrition apps are generally good at detecting individual Western dishes when they are served separately on a plate,” says Dr Chen of the Charles Perkins Center, “but they often struggle with mixed dishes such as spaghetti Bolognese or a hamburger.”
“This issue is common with Asian cuisines, which typically contain a variety of mixed ingredients that are not in the respective apps' databases, which can lead to errors when calculating the energy content of a particular meal.”
Going forward, the study recommends several steps to improve nutrition apps, including ensuring that the educational content and advice they provide is evidence-based and trustworthy, which can be achieved through collaboration with nutrition experts.
“To make nutrition apps more reliable and accurate, developers should involve nutritionists in their development, train AI models with diverse food images (especially mixed and culturally diverse dishes), expand food ingredient databases, and educate users on how to take high-quality food images for better recognition accuracy,” said Dr Chen.
“If you're monitoring your health, such as managing high blood pressure or tracking your sodium intake, it's important to compare your food choices with nutrition labels and/or consult with a certified, practicing dietitian. A dietitian's expertise is invaluable in these cases, as they can provide a more accurate estimate of how much energy your body is consuming and how much energy you most need to achieve an overall healthy diet.”
The evaluation was conducted using the Mobile App Rating Scale (MARS) and the App Behavior Change Scale (ABACUS).
As a result of the evaluation, Noom received an average score of 4.44 out of 5 on the MARS scale, meaning it received very high marks in engagement, functionality, aesthetics, and quality of information. It also received a perfect score of 21/21 on ABACUS, due to the many features it incorporates that encourage behavior change, goal setting, tracking, and educational content.
Among other AI-powered apps, “MyFitnessPal” and “Fastic” successfully recognized a sample of 22 images of various foods and drinks, achieving success rates of 97% and 92%, respectively.
More information:
Xinyi Li et al. “Evaluating the quality and comparative validity of manual food recording and artificial intelligence-based food image recognition in a nutritional care app” Nutrients (2024). DOI: 10.3390/nu16152573
Provided by University of Sydney
Citation: AI food tracking apps need to be improved for accuracy and cultural diversity, study says (August 29, 2024) Retrieved August 29, 2024 from https://medicalxpress.com/news/2024-08-ai-food-tracking-apps-accuracy.html
This document is subject to copyright. It may not be reproduced without written permission, except for fair dealing for the purposes of personal study or research. The content is provided for informational purposes only.