Validity and accuracy of artificial intelligence-based dietary intake assessment methods: a systematic review

Area of Science:

• Nutritional Epidemiology
• Artificial Intelligence in Health
• Dietary Assessment Methods

Background:

• Accurate dietary data is crucial for linking food intake to health outcomes in nutritional epidemiology.
• Traditional dietary assessment methods often face challenges in accuracy and validity.
• Artificial intelligence (AI) offers advanced statistical models for improved nutrient and food analysis.

Purpose of the Study:

• To systematically review the validity and accuracy of AI-based dietary intake assessment (AI-DIA) methods.
• To synthesize evidence on the performance of AI-DIA compared to traditional assessment techniques.
• To identify research gaps and future directions for AI-DIA in nutritional studies.

Main Methods:

• Systematic literature search conducted across EMBASE, PubMed, Scopus, and Web of Science databases.
• Inclusion of 13 studies meeting PRISMA guidelines, with data analyzed up to December 2024.
• Analysis of AI techniques used, study settings (preclinical), and reported correlation coefficients for nutrient estimations.

Main Results:

• AI-DIA methods demonstrated promising accuracy, with correlation coefficients over 0.7 reported for calories, macronutrients, and micronutrients in multiple studies.
• Deep learning (46.2%) and machine learning (15.3%) were the predominant AI techniques employed.
• A moderate risk of bias was observed in 61.5% of the studies, with confounding bias being most common.

Conclusions:

• AI-DIA methods represent reliable and valid alternatives for estimating nutrient and food intake.
• Further research with diverse populations, larger sample sizes, and robust experimental designs is necessary to solidify AI-DIA's role.
• AI-DIA has the potential to significantly advance nutritional epidemiology research.