Washington, July 27 (SANA) Artificial intelligence-powered calorie-tracking apps may substantially underestimate the calorie and fat content of meals, potentially affecting the accuracy of diet monitoring, according to a new study by researchers at the U.S. National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK), part of the National Institutes of Health (NIH).
According to a report published by ScienceDaily on Sunday, researchers evaluated four widely used AI-based food image analysis apps—MyFitnessPal, Lose It!, CalAI and Appediet—using photographs of 102 meals prepared in a tightly controlled research kitchen, where every ingredient was weighed to an accuracy of 0.1 grams.
The study found that the apps underestimated calorie content by approximately 250 to 345 calories per meal. They also underestimated fat content by about 30 grams on average, while carbohydrate estimates were generally more accurate than estimates for other nutrients.
Preliminary findings from a second phase of the research, involving more than 200 additional meals, indicated that the apps were less accurate when assessing low-carbohydrate, high-fat meals, such as ketogenic diets, increasing the likelihood of underestimating their actual nutritional value.
Researchers said that combining AI-powered food image analysis with traditional dietary assessment methods, including manually entering meal ingredients and portion sizes, could improve the accuracy of calorie tracking in everyday use. They noted that the findings remain preliminary.
The study highlights the need for caution when relying on AI-based nutrition apps until more accurate and reliable tools become available to support dietary monitoring.
R.H/M.F
