JMIR Publications

ABSTRACT

Background:

There is growing interest in the use of ecological momentary assessments to collect data on health behaviors and their predictors on a micro-timescale (at different points within or across days) but documenting eating behaviors remains a challenge.

Objective:

The goal of this pilot study (n=48) was to examine the feasibility of using smartphone-captured images to document eating behaviors in real time.

Methods:

Participants completed the Block Fat/Sugar/Fruit/Vegetable Screener to provide a measure of their eating behaviors throughout a typical week. Participants then took pictures of their meals and snacks and answered brief survey questions for seven consecutive days using a commercially available smartphone application. Participant acceptability was determined through a questionnaire about their experiences, administered at the end of the study. Meal/snack images were uploaded to Amazon’s Mechanical Turk (mTurk), a crowdsourcing distributed human intelligence platform where two workers assigned a count of certain food categories to the images (fruits, vegetables, salty snacks, and sweet snacks). Agreement among mTurk workers was assessed, and weekly counts were calculated based on the food images and compared with reports from the Block Screener.

Results:

Participants reported little difficulty uploading photos and they remembered to take photos most of the time. Images classified by mTurk workers (n=1014) showed moderate agreement for vegetables (67.9%) and high agreement for all other food categories (85.4% for fruits, 83.2% for salty snacks, and 81.4% for sweet snacks). There were no significant differences between weekly consumption frequency for any food category as recorded in the food images compared with the Block Screener, suggesting this approach may accurately reflect typical eating behaviors.

Conclusions:

In summary, our approach offers a potential time-efficient and cost-effective strategy for capturing eating behaviors in real time.