Summary: A recent study showed that AI-based calorie-tracking apps that estimate meal content from photos may undercount the actual calories by about 345 calories and about 30 grams of fat per meal. This underestimation may affect the accuracy of diet management, especially when relying completely on these apps. The findings point to the importance of integrating these tools with other traditional methods to ensure a more accurate assessment of calories and nutrients, given their impact on public health and weight control.
🧬 How do AI-based calorie-tracking apps work?
These apps rely mainly on artificial intelligence (AI) technologies and image-recognition algorithms to identify food types and portion sizes in the captured image. After analyzing the image, the apps use nutrition databases to determine calorie content and various nutrients.
This approach offers a quick and convenient alternative to manually entering food data, attracting a large group of users who aim to monitor their health or lose weight.
However, researchers indicate that the accuracy of these apps has not been sufficiently evaluated so far, raising questions about the reliability of the estimates they provide.
🧪 A precise test using carefully measured meals
A research team from the U.S. National Institutes of Health used meals prepared in an accurate metabolic kitchen, where the ingredients of each meal were measured with precision down to 0.1 gram. This made it possible to conduct a precise comparison between the actual nutritional content and the apps’ estimates.
A total of 102 meals were photographed in a study that included two different diets (the ketogenic diet and the regular diet). The images were then submitted to four popular apps: MyFitnessPal, LoseIt!, CalAI and Appediet to assess how closely these apps’ estimates matched the real nutritional values.
This advanced method provided preliminary data for testing the apps’ reliability under strict conditions.
🩺 Study results: estimates below reality
The comparison of estimates showed that the apps reduced the calorie content by about 250 to 345 calories on average per meal.
There were also reports of the estimated fat amount being reduced by about 30 grams per meal, indicating a major challenge specifically in fat assessment.
- MyFitnessPal and LoseIt! showed better accuracy with high-calorie meals compared with lower-calorie meals.
- All apps showed more accurate estimates for carbohydrates than for protein or fat.
- Users who do not adjust portion-size estimates or enter additional details should be careful about relying completely on these apps.
According to the researchers, part of this error is due to the software’s difficulty in estimating food categories that contain high fat proportions.
🌱 Challenges of the ketogenic diet in AI-based estimation
The researchers carried out additional analyses of more than 200 meals belonging to low-carbohydrate, high-fat diets, such as the ketogenic diet.
The initial results showed that these apps have greater difficulty evaluating keto meals because of the weak accuracy in estimating the high fat content in them.
This highlights the need to improve AI algorithms so they become more sensitive to diverse dietary patterns, especially those that rely on fat-based energy sources.
🧠 Integrating AI technologies with traditional assessment methods
The researchers recommended using AI-powered calorie-tracking tools alongside other traditional methods such as written food logging or manual portion measurement.
Such a combination can improve estimate accuracy and provide individuals with better nutritional care.
At the same time, these apps are useful for providing a quick general picture of food intake, but they are not a complete replacement for precise monitoring methods in research or clinical settings.
Conclusion:
- Relying entirely on AI estimates may lead to underestimating the true value of estimated calories and fat.
- Keto meals pose a greater challenge for these apps in terms of accuracy.
- The integration of AI and traditional methods is the best solution for achieving reliable results.
📚 Related articles
Discover more from Mohdbali
Subscribe to get the latest posts sent to your email.
