Validation of artificial intelligence-based application to estimate nutrients in daily meals

Area of Science:

• Nutrition Science
• Artificial Intelligence
• Health Technology

Background:

• Diet modification is crucial for managing metabolic syndrome and reducing cardiovascular disease risk.
• Accurate quantification of essential nutrients in daily meals is a significant challenge.
• HAKARIUM, an AI-based application, estimates nutrient components from photographs, but its real-world applicability is unproven.

Purpose of the Study:

• To evaluate the accuracy of the HAKARIUM AI application in estimating nutrient content (calories, protein, salt) in real-world meals.
• To compare AI-estimated nutrient values with officially calculated values from a university cooperative society.

Main Methods:

• Analysis of 62 lunchtime meals served at a university cooperative society.
• Estimation of energy (kcal), protein (g), and salt (g) using the HAKARIUM AI application.
• Comparison of AI-estimated nutrient values against official calculations provided by the university.

Main Results:

• AI-estimated energy: 636 (493, 835) kcal vs. actual: 382 (358, 431) kcal.
• AI-estimated protein: 25.7 (19.7, 36.3) g vs. actual: 17.1 (13.9, 18.9) g.
• AI-estimated salt: 4.2 (3.5, 4.6) g vs. actual: 2.9 (2.6, 3.1) g.
• Significant correlations (r > 0.80, p < 0.05) and good agreement were observed between AI and actual values for all nutrients.

Conclusions:

• AI-based nutrient estimation demonstrates relatively good agreement with actual calculated values.
• The HAKARIUM application shows potential for practical use in estimating meal nutrient content.