Validity of an AI-based Program to Identify Foods and Estimate Food Portion Size (PortionSizeAI)

Testing the Validity of an Artificial Intelligence-based Program to Identify Foods and Estimate Food Portion Size Among Adults, a Pilot Study

The purpose of this study is to test the accuracy of the Nutrition Artificial Intelligence in the Openfit app during meals in a controlled laboratory setting

Study Overview

• Status: Completed
• Conditions: Nutrition Assessment
• Intervention / Treatment: Device: PortionSize AI

Detailed Description

For this pilot study, using a convenience sample, the investigators will recruit up to 25 adults to use the Nutrition AI technology in Openfit to identify and estimate portion size of foods plated in a laboratory setting at Pennington Biomedical Research Center (PBRC) and/or Louisiana State University (LSU). Laboratory members within the Ingestive Behavioral Laboratory will also test the ability of Nutrition AI to identify foods and to quantify foods provided in the laboratory. Meals will be simulated, and participants will not consume the foods provided.

• Study Type: Interventional
• Enrollment (Actual): 24
• Phase: Not Applicable

Contacts and Locations

Study Locations

• United States
  • Louisiana
    • Baton Rouge, Louisiana, United States, 70808
      • Pennington Biomedical Research Center

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years to 62 years (Adult)
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

• Male or female
• Aged 18-62 years
• Self-reported body mass index (BMI) 18.5-50 kg/m2

Exclusion Criteria:

• Any condition or circumstance that could impede study completion
• Unfamiliar with or not able to use an iPhone

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Other
• Allocation: N/A
• Interventional Model: Single Group Assignment
• Masking: None (Open Label)

Number of Arms

1

Arms and Interventions

Participant Group / Arm

Intervention / Treatment

Experimental: Experimental; Training and use of Openfit; Using the app to estimate food intake from simulated meals in a laboratory at PBRC or LSU (participants will not eat food during the meals); Rating the usability and satisfaction of the app

Device: PortionSize AI; For this pilot study, using a convenience sample, the investigators will recruit up to 25 adults to use the Nutrition AI technology in Openfit to identify and estimate portion size of foods provided in a laboratory setting at Pennington Biomedical Research Center (PBRC) and/or Louisiana State University (LSU). Laboratory members within the Ingestive Behavioral Laboratory will also test the ability of Nutrition AI to identify foods and to quantify foods provided in the laboratory. Meals will be simulated, and participants will not consume the foods provided.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Identification of Food Plated Using the Openfit Mobile App	Agreement surrounding identification of food and beverages provided compared with known identification, at the item level, and across all items where identification is determined by: 1) Nutrition AI without correction (automated), 2) Nutrition AI with user correction (semi-automated) For a food identified through the Nutrition AI to be considered an exact food match, the name of the food identified must match or be a close match to the food served. For example, a fruit cocktail identified as a fruit salad is an acceptable match. Proportions will be used to assess whether the percentage of food items plated that were correctly identified by Nutrition AI is different to the percentage of foods correctly identified by a criterion method (human rater). Descriptive data will also be used to describe the frequency at which food plated was correctly identified for all food items across all participants. In total there was 255 food items tested across all participants.	One study visit of ~2 hours
Portion Size Estimation (kcal) of Food Plated Using the Openfit Mobile App	Error between mean estimates of food plated (kcal) and known food plated (kcal), determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) Mean error and Bland-Altman analysis will be performed to determine errors in estimation of food plated from the Nutrition AI compared to estimations from the criterion measure (weighed food).	One study visit of ~2 hours
User Satisfaction of the Openfit Mobile App for Recording Food Plated	After completing assessment of food plated, participants will complete a user satisfaction survey (USS). The USS was adapted from a previous version used to assess the usability of a mobile application for dietary assessment. The USS includes five quantitative questions and three open response questions. The quantitative questions will each be scored using a 6-point Likert scale, with 1 being the lowest and worst score, and 6 being the highest and best score. Data for each of the five quantitative responses in the USS will be averaged across participants and presented separately as mean (SD). Open responses will be evaluated using qualitative methods to identify common themes.	One study visit of ~2 hours
Usability of the Openfit Mobile App for Recording Food Plated	Participants will complete the Computer Usability Satisfaction Questionnaire (CSUQ). The CSUQ is frequently used to assess the usability of mobile applications. The CSUQ consists of 19 questions, each scored using a 7-point Likert scale (with 1 being the lowest and best score and 7 being the highest and worst score) and participants will rate satisfaction, usefulness, information quality, and interface quality of the Openfit app. The average of these 19 questions (1 being the best average score and 7 being the worst average score) provides an overall usability score.	One study visit of ~2 hours

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Identification of plate waste using the Openfit mobile app	Agreement surrounding identification of plate waste compared with known identification, at the item level, and across all items, where identification is determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) For a food identified through the Nutrition AI to be considered an exact food match, the name of the food identified must match or be a close match to the food served. For example, a fruit cocktail identified as a fruit salad is an acceptable match. Proportions will be used to assess whether the percentage of food items that were correctly identified by Nutrition AI is different to the percentage of foods correctly identified by a criterion method (human rater). Descriptive data will also be used to describe the frequency at which plate waste was correctly identified.	One study visit of ~2 hours
Portion size estimation (gram weight) of plate waste using the Openfit mobile app	Error between mean estimates of portion weights (gram weight) and known portion weights (gram weight), where estimates of plate waste are determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) Mean error and Bland-Altman analysis will be performed to determine errors in estimation of plate waste from the Nutrition AI compared to estimations from the criterion measure (weighed food).	One study visit of ~2 hours
Portion size estimation (kcal) of plate waste using the Openfit mobile app	Error between mean estimates of portion size (kcal) and known portion size (kcal), where estimates of plate waste are determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) Mean error and Bland-Altman analysis will be performed to determine errors in estimation of plate waste from the Nutrition AI compared to estimations from the criterion measure (weighed food).	One study visit of ~2 hours
User Satisfaction of the Openfit mobile app for recording plate waste	After completing assessment of plate waste, participants will complete a user satisfaction survey. The investigators adapted a 10-question user satisfaction survey that has been administered in prior studies to quantify satisfaction, ease of use, and adequacy of training for mobile apps. Responses to the user satisfaction questionnaire will be assessed using frequencies and percentages for Likert scale data. Open responses will be evaluated using qualitative methods to identify common themes.	One study visit of ~2 hours
Portion size estimation (gram weight) of food intake using the Openfit mobile app	Error between mean estimates of portion weights (gram weight) and known portion weights (gram weight), where estimates of food intake (food plated - plate waste) are determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) Mean error and Bland-Altman analysis will be performed to determine errors in estimation of food intake from the Nutrition AI compared to estimations from the criterion measure (weighed food).	One study visit of ~2 hours
Portion size estimation (kcal) of food intake using the Openfit mobile app	Error between mean estimates of food intake (kcal) and known food intake (kcal), where estimates of food intake (food plated - plate waste) are determined by: 1) Nutrition AI without user correction (automated), 2) Nutrition AI with user correction (semi-automated) Mean error and Bland-Altman analysis will be performed to determine errors in estimation of food intake from the Nutrition AI compared to estimations from the criterion measure (weighed food).	One study visit of ~2 hours

Collaborators and Investigators

• Sponsor: Pennington Biomedical Research Center
• Collaborators: National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK)
• Investigators: Principal Investigator: Chloe P Lozano, PhD, Pennington Biomedical Research Center

Study record dates

Study Major Dates

• Study Start (Actual): April 27, 2022
• Primary Completion (Actual): June 3, 2022
• Study Completion (Actual): June 3, 2022

Study Registration Dates

• First Submitted: April 12, 2022
• First Submitted That Met QC Criteria: April 18, 2022
• First Posted (Actual): April 25, 2022

Study Record Updates

• Last Update Posted (Estimated): November 16, 2023
• Last Update Submitted That Met QC Criteria: November 14, 2023
• Last Verified: November 1, 2023

More Information

Terms related to this study

• Keywords: Nutrition Assessment
• Other Study ID Numbers: PBRC 2022-011; T32DK064584 (U.S. NIH Grant/Contract)

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Any identifiers might be removed from the participants identifiable information and after such removal, the information could be used for future research studies or given to another investigator for future research without additional informed consent from the subject or legally authorized representative.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No