Related Experiment Video
Updated: Jan 13, 2026
Iterative Development of an Innovative Smartphone-Based Dietary Assessment Tool: Traqq
Published on: March 19, 2021
Measuring diet intake in adolescents: Relative validation of an artificial intelligence enhanced, image assisted
Audrey Moyen1, Antonio Rossi2, Maurya Hart2
1Department of Nutritional Sciences, University of Toronto, Toronto, ON, Canada; School of Human Nutrition, McGill University, Montreal, QC, Canada; Research Institute of the McGill University Health Centre, Montreal, QC, Canada.
Background & Aims:
Puberty is a critical period of development during which nutritional exposures are known to shape long-term health and the risk of chronic diseases. Current dietary assessment methods have limitations for use in large cohorts of adolescent populations. We aimed to evaluate the relative validity of Keenoa (not an acronym), an artificial intelligence-enhanced image-assisted mobile application, against the validated Automated Self-Administered 24 h recall (ASA24)-Canada web-based platform, among adolescents in the CHILD Cohort Study.
Methods:
Using a randomized crossover design, participants aged 11-15 years old completed three days (two weekdays and one weekend day) of both Keenoa food tracking and ASA24 food recalls. Differences in reported intakes were analyzed using paired t-tests or Wilcoxon signed-rank test and deattenuated correlations by Spearman's coefficient. Agreement and bias were determined using Bland-Altman's test, and inter-quartile cross-classification agreement was assessed using weighted Cohen kappa.
Results:
This study included 141 participants with a mean age of 12.2 ± 0.8 years; of them 74 (52.5 %) males; and 88 (62.4 %) identified as Caucasian/White. Mean ± SD reported energy intakes (kcal/d) were 1976 ± 451 and 1978 ± 425, with ASA24 and Keenoa, respectively (P = 0.95). Mean reported macronutrient, iron, and potassium intakes did not significantly differ between tools. Reported fiber intake was higher, while sodium, calcium and vitamin D intakes were lower with Keenoa compared to ASA24 (P values < 0.001-0.025). Deattenuated correlations between tools ranged from r = 0.77 to 1.00 (all p< 0.01) and weighted Cohen κ scores ranged from 0.22 to 0.42 (all p < 0.001). Among all participants, 121 (85.8 %) and 78 (55.3 %) completed all 3 requested days with Keenoa and ASA24, respectively (P< 0.01).
Conclusion:
The artificial intelligence-enhanced image-assisted Keenoa mobile application showed strong to moderate relative validity against ASA24 for energy, macronutrient, potassium and iron intakes. Vitamin D, calcium, fiber and sodium showed limited relative agreement based on mean differences. This novel tool may facilitate dietary assessment and reduce attrition bias in cohort studies. Future validation using objective biomarker measures will help establish true validity.
