A recent study shows that better training of artificial intelligence (AI) is necessary for creating nutrition apps that help users monitor food intake and manage weight.
A recent study from the University of Sydney highlights the need for enhanced training in artificial intelligence (AI) for the development of nutrition apps that assist in tracking food consumption and managing weight.
Initially, researchers evaluated 800 different apps and narrowed down their selection to 18 for an in-depth assessment. These included apps that utilized AI as well as manual food-logging apps, which were evaluated for their capability to identify ingredients and calculate energy content.
Dr. Juliana Chen, the primary author and a registered dietitian, lecturer, and researcher in Nutrition and Dietetics at the University of Sydney, mentions that while AI-based apps provide conveniences that manual apps do not, they still require careful usage.
“For many people, using apps to log food intake or manage weight can sometimes be overwhelming,” Dr. Chen stated. “Incorporating AI features, such as food image recognition, could streamline this process quite a bit.”
“Nevertheless, it’s essential to confirm that the portion size recognized by the app corresponds with the amount you’ve actually consumed. Some apps merely identify the food, while others also attempt to estimate the portion size and energy intake. Therefore, for those who are trying to lose weight, it’s vital to ensure that the app’s estimations match your actual consumption,” she advised.
A significant focus of the study was to assess the accuracy and adaptability of these apps across three different dietary guidelines: Western, Asian, and the recommended diet based on the Australian Dietary Guidelines, to ensure cultural dietary preferences were adequately represented.
Under Dr. Chen’s guidance, Master of Nutrition and Dietetics students Xinyi Li, Annabelle Yin, and Ha Young Choi discovered that manual food-logging apps tended to overstate energy intake for the Western diet by approximately 1040 kilojoules, while they underestimated energy intake for the Asian diet by around 1520 kilojoules and the recommended diet by about 944 kilojoules.
On the other hand, AI-driven apps struggled with accurately gauging the energy content of complex Asian dishes; for instance, the calorie count for beef pho was overestimated by 49 percent, and pearl milk tea had calorie counts underestimated by as much as 76 percent.
“Typically, nutrition apps that utilize AI perform better at identifying separate Western dishes when arranged individually on a plate,” Dr. Chen, who is also affiliated with the Charles Perkins Centre, explained. “However, they find it challenging to assess mixed dishes like spaghetti bolognese or hamburgers. This issue is particularly prevalent with Asian foods, which often consist of multiple components not represented in the app’s database, leading to potential miscalculations in energy content.”
Looking ahead, the study suggests several improvements for nutrition apps. These include making sure that the educational information and advice provided are based on sound research and collaborating with nutrition specialists to achieve this.
“To bolster the reliability and precision of nutrition apps, developers need to involve dietitians in their conception, train AI systems on a variety of food images—especially for mixed and culturally diverse dishes—expand food composition databases, and educate users about taking high-quality food images for better recognition,” Dr. Chen emphasized.
“If you’re tracking your health, whether it’s managing hypertension or monitoring your sodium intake, it’s critical to compare your dietary choices with nutrition labels or confer with a registered dietitian. The insights provided by a dietitian can be crucial, as they offer more accurate evaluations of your energy intake and dietary needs for achieving a well-balanced diet.”
This evaluation was carried out using the Mobile App Rating Scale (MARS) and the App Behaviour Change Scale (ABACUS).
After the review, “Noom” scored an impressive 4.44 out of 5 on the MARS scale, indicating high ratings related to engagement, functionality, design, and quality of information. It also earned a perfect score of 21/21 on the ABACUS scale for incorporating numerous features that encourage behavior modification, goal setting, tracking, and educational resources.
Other AI-enabled apps like “MyFitnessPal” and “Fastic” successfully identified a set of 22 food and beverage images, achieving recognition rates of 97 percent and 92 percent, respectively.
