AI food tracking apps need improvement to address accuracy and cultural diversity, says study

A brand new University of Sydney research has discovered that improved synthetic intelligence (AI) coaching is required when growing diet apps which are used to trace meals consumption or handle weight.

Researchers initially screened 800 apps earlier than deciding on 18 for additional analysis. These 18 apps, which included each AI-integrated and handbook food-logging diet apps, had been then assessed for his or her capacity to acknowledge components and estimate power content material.

The findings are revealed within the journal Nutrients.

Dr. Juliana Chen, lead writer of the research and accredited training dietitian, lecturer and researcher within the Discipline of Nutrition and Dietetics on the University of Sydney, means that whereas AI-integrated apps provide comfort over handbook food-logging, they need to be used rigorously.

“When sufferers or the general public use apps to trace meals consumption or handle weight, the method can usually really feel burdensome,” Dr. Chen stated. “Adding AI options like meals picture recognition may make the method a lot simpler for everybody.

“However, you will need to all the time double-check that the portion dimension detected matches what you ate. Some apps solely determine the meals, whereas others additionally estimate portion dimension and power consumption. So, for these present process weight reduction, it’s essential to confirm that the app’s estimates align with what you could have eaten.”

A key a part of the research was checking how correct and adaptable these apps had been throughout three completely different weight loss program plans—Western, Asian, and really helpful (primarily based on the Australian Dietary Guidelines)—to make sure a variety of cultural meals preferences had been thought of.

Under Dr. Chen’s supervision, Master of Nutrition and Dietetics college students Xinyi Li, Annabelle Yin and Ha Young Choi discovered that handbook food-logging apps overestimated power consumption for the Western weight loss program by a mean of 1,040 kilojoules, whereas they underestimated power consumption for the Asian weight loss program and the really helpful weight loss program by a mean of 1,520 kilojoules and 944 kilojoules respectively.

In distinction, AI-integrated meals apps usually had issue precisely figuring out power content material for combined Asian dishes, for instance, the energy for beef pho had been overestimated by 49%, whereas pearl milk tea had calorie underestimations of as much as 76%.

“Nutrition apps with AI-integration are usually higher at detecting particular person Western meals when they’re separated on a plate,” stated Dr. Chen who can be from the Charles Perkins Center. “However, they usually battle with combined dishes, equivalent to spaghetti bolognese or hamburgers.

“This concern is extra widespread with Asian dishes, which often comprise quite a lot of combined elements that will not be discovered within the respective apps database, resulting in attainable errors when calculating the power quantity of a specific meal.”

Moving ahead, the research recommends a number of steps for the advance of diet apps. This consists of making certain that the tutorial content material and recommendation supplied by the apps are evidence-based and reliable, which may be achieved by way of collaboration with diet consultants.

“To improve the credibility and accuracy of diet apps, creators ought to have interaction dietitians of their growth, prepare AI fashions with numerous meals photographs—notably for combined and culturally diverse dishes—increase meals composition databases and educate customers on capturing high-quality meals photographs for higher recognition accuracy,” stated Dr. Chen.

“If you are monitoring your well being, equivalent to managing hypertension or monitoring your sodium consumption, it is vital to match your meals decisions with diet labels or seek the advice of with an accredited training dietitian. A dietitian’s experience is invaluable in these instances, as they’ll present extra correct estimates of how a lot power your physique is consuming and what it requires most to attain a holistically nutritious diet.”

This evaluation was performed utilizing the Mobile App Rating Scale (MARS) and the App Behavior Change Scale (ABACUS).

Following the analysis, “Noom” acquired a mean rating of 4.44 out of 5 on the MARS scale, that means it was rated very extremely when it comes to engagement, performance, aesthetics, and knowledge high quality. It additionally acquired an ideal 21/21 ABACUS ranking for incorporating many options that promoted behavioral change, aim setting, monitoring and academic content material.

Among the opposite AI-powered apps, “MyFitnessPal” and “Fastic” efficiently acknowledged a pattern of twenty-two photographs of varied meals and drinks, reaching success charges of 97% and 92%, respectively.