MealMeter: Using Multimodal Sensing and Machine Learning for Automatically Estimating Nutrition Intake.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Nutritional Science

Background:

• Accurate macronutrient estimation is crucial for precision nutrition, metabolic health monitoring, and glycemic control.
• Traditional dietary assessment methods are time-consuming and prone to inaccuracies.
• Existing AI-driven frameworks often require extensive data.

Purpose of the Study:

• To develop and evaluate MealMeter, a novel machine learning-driven method for estimating meal macronutrient composition.
• To leverage multimodal sensor data from wearable and mobile devices for enhanced dietary assessment.
• To improve the accuracy and automation of meal tracking for personalized nutrition.

Main Methods:

• Collected multimodal sensor data (physiological signals, motion data, environmental cues) from 12 participants.
• Developed lightweight machine learning models trained on labeled meal events.
• Integrated physiological signals (continuous glucose, heart rate variability) and inertial motion data.

Main Results:

• MealMeter accurately predicted carbohydrate, protein, and fat content.
• Achieved low average mean absolute error (MAE) of 13.2 grams for carbohydrates.
• Demonstrated significant improvement over baseline methods, including foundation models.

Conclusions:

• Multimodal sensing combined with machine learning significantly enhances meal macronutrient estimation.
• MealMeter offers a potential solution for automated meal tracking and personalized dietary interventions.
• The system supports precision nutrition and management of metabolic disorders like diabetes and obesity.