The research, published in the journal Nutrients and funded by the Ministry of Science and ICT Korea, stressed how the findings underscore the breadth of AI applications within nutrition.
“From these applications, we observed a paradigm shift in how nutritional research offers innovation tools and gives insights that hold promising potential in improving individual health outcomes and advancing public health initiatives,” the Korean researchers noted.
Mariette Abrahams, CEO and founder of specialized nutrition innovation consultancy and platform Qina stressed that while the findings are positive, there is a lot of work still to be done to ensure effective implementation of AI.
“We are currently still grappling with longstanding challenges of incomplete food databases, a lack of insights into detailed nutrient composition and the historic exclusion of non-western diets,” she told NutraIngredients.
“These limitations greatly affect the quality and accuracy of any dietary and meal analysis, and the subsequent recommendations we can make."
When it comes to disease prediction and management, social determinants have a big impact on health, yet are often omitted in AI training datasets, she added.
“We also know from recent research that understanding the psychological and behavioral profiles is important for developing personalized nutrition recommendations. However, until we have more and better research into this area, it will be difficult to develop quality AI systems.”
Nutrition and AI
Optimal nutrition is vital for overall health and wellbeing and is known to prevent and manage non-communicable diseases and a range of health issues such as obesity, diabetes and heart disease. Nutritional therapy is rapidly growing in prevalence due to the rise of personalized nutrition and genomics, allowing for highly individualized advice to more accurately improve health.
As nutrition has traditionally relied on observational studies and clinical trials, the use of AI technologies has the potential to uncover complex relationships in large datasets, identify patterns and generate actionable information to further advance the field of personalized nutrition. Thus, the present study sought to gather existing evidence on the use of AI to address nutritional challenges.
Smart and personalized nutrition
The researchers collected a total of 31 relevant papers through a systematic literature review of the PubMed, IEEE and Google Scholar online databases to obtain for inclusion.
The results regarding the application of AI, machine learning and deep learning within nutrition were classified into five main clusters: smart and personalized nutrition, dietary assessment, food recognition and tracking, predictive modeling for disease, and disease diagnosis and monitoring.
Multiple papers observed significant potential for the integration of AI techniques into personalized nutrition. It was identified that AI could be used to create nutritional recommendations depending on context, taking into account factors such as addressing nutritional insecurity and recommending cooking methods to optimize nutrition.
In addition, AI enabled recipe and overall household nutrition analysis to provide personalized nutrition recommendations to optimize health. AI also helps personalized recommendations account for certain health conditions and food preferences.
Machine learning and deep learning were shown to have potential within precision nutrition and metabolomics to advance and optimise research.
Dietary assessment, food recognition and tracking
The researchers identified the ability of AI to automate nutrient analysis from dietary records, food images and other sources, reducing the bias associated with self-reporting within nutritional research.
One developed system called goFOODTM was found to accurately estimate a meal’s calorie and macronutrient content based on food images captured on a smartphone.
A further deep learning method was proposed called OptmWave, which uses learning-based, near-infrared hyperspectral imaging to predict nutritional content of food.
Predictive modeling for disease
AI also demonstrated potential to improve prediction and treatment of diseases, care and medication of patients, while monitoring patients in real time. In addition, it was noted that such technologies would benefit healthcare practices with faster diagnoses, personalized medicines, disease evaluation and monitoring, and reduced healthcare costs.
Disease diagnosis and monitoring
It was reported that machine learning could be applied to personalized nutrition and health biomarker optimization for disease diagnosis, with the use of metabolomics to study unique chemical fingerprints left by cellular processes. In addition, it was proposed that a deep learning system could classify BMI based on biochemistry profile to enable accurate personalized nutrition recommendations for disease management.
“All these findings collectively show the transformative possibility of AI in reforming the field of nutrition research and promoting personalized approaches to health and well-being,” the report concluded.
Abrahams stressed that healthcare professionals and domain experts should be actively involved in the development of AI systems, while more multidisciplinary research with a view to sustainable behavior change through AI was required.
Source: Nutrients
“Applications of Artificial Intelligence, Machine Learning, and Deep Learning in Nutrition: A Systematic Review”
Authors: Tagne Poupi Theodore Armand et al.