Scalable multimodal infrastructure for passive dietary monitoring using smartwatches

Background

Traditional dietary assessment methods are limited by recall bias and participant burden. Wearable technologies enable passive monitoring of eating behaviour, but scalable data acquisition systems remain a key barrier.

Methods

We developed and deployed a multimodal infrastructure within the FITMATE pilot to enable continuous dietary monitoring using consumer smartwatches. The system integrates: (1) a custom Garmin smartwatch application for raw sensor acquisition; (2) a REDCap-based electronic data capture platform linked to MyCap for dietary diaries and meal images; (3) centralized data collection via the Garmin Health API; and (4) a custom dashboard for device and study management.

Results

Over 100 participants have been enrolled, generating synchronized multimodal data streams including wearable sensor data, meal photographs, and self-reported dietary records. The infrastructure demonstrated stable performance across all components, supporting continuous and concurrent data acquisition. Its modular design enabled iterative updates, including SDK compatibility improvements, without interrupting data collection.

Conclusions

This scalable infrastructure enables integration of passive wearable sensing with active dietary reporting, addressing key challenges in large-scale nutritional monitoring. It provides a robust foundation for developing machine learning models for automated eating detection and caloric estimation in real-world settings.

Key messages

• A scalable infrastructure enables continuous dietary monitoring by integrating smartwatch sensing, mobile data capture, and centralized data management.

• Multimodal real-world data collection supports development of AI models for automated eating detection and nutrition assessment.