Evaluating a Multitask Artificial Intelligence Model Compared With Humans for Portion-Size Estimation.

Reliable dietary data for adolescents in low- and middle-income countries (LMICs) are limited due to high costs and estimation errors in traditional dietary assessment methods. Although technology-assisted dietary assessment tools are becoming popular, few have been validated in LMICs. This study validated the PlantVillage Food Recognition Assistance and Nudging Insights (FRANI), an artificial intelligence-assisted mobile application for dietary assessment, against weighed food records (WFR) and multipass 24-h recalls (24HR) among adolescent girls aged 14‒18 y (n = 60) in urban/semi-urban communities in Sri Lanka. Dietary intake was assessed over 2 non-consecutive days using 3 methods: FRANI, WFR, and 24HR. The equivalence of nutrient intake was evaluated using mixed-effect models accounting for repeated measures by comparing intake ratios (FRANI/WFR and 24HR/WFR) with 10%, 15%, and 20% equivalence bounds. The concordance correlation coefficient was utilized to assess the agreement between methods. FRANI demonstrated equivalence with WFR at the 10% bound for energy and vitamin A; 15% for protein, fiber, iron, and zinc; and 20% for fat, niacin, and folate intakes. Comparisons between 24HR and WFR found that no nutrients fell within the 10% bound. Energy, protein, fat, iron, niacin, and vitamin A intakes were equivalent at 15% bound, whereas fiber, calcium, folate, and vitamin C intakes were equivalent at 20% bound. Concordance correlation coefficient ranged from 0.49 to 0.89 for FRANI compared to WFR, and 0.44 to 0.84 for 24HR compared with WFR. Omission errors were 2% for FRANI and 12% for 24HR, and intrusion errors were 7% and 9%, respectively. PlantVillage FRANI application accurately estimated nutrient intakes of adolescent girls in Sri Lanka compared to the WFR. Its performance was at least comparable to the traditional 24HR method, supporting its potential as a scalable alternative for dietary assessment in similar LMIC populations.

Background: Effective glycemic control in diabetes management relies heavily on dietary carbohydrate knowledge. This study aimed to assess carbohydrate knowledge in individuals with type 1 diabetes (T1D) and insulin-treated type 2 diabetes (itT2D) using the GluciQuizz tool. Methods: A total of 465 persons (96 with T1D, 153 with itT2D; 89 and 127 matched controls without diabetes, respectively) from the French NutriNet-Santé prospective cohort were included. Participants completed the GluciQuizz questionnaire, which evaluates carbohydrate knowledge across five domains: carbohydrate food recognition; carbohydrate food content; nutrition label reading; glycemic targets and hypoglycemia prevention and treatment; and carbohydrate content of meals. Results: The mean age ± standard deviation of participants with diabetes was 65.8 ± 11.2 years, 44.2% male, with a diabetes duration of 23.3 ± 12.9 years. T1D participants scored significantly higher on the GluciQuizz compared to those with itT2D (23.9 ± 5.0 vs. 17.5 ± 5.6, p < 0.001). In secondary analysis, T1D participants showed superior knowledge to their matched controls without diabetes, whereas itT2D participants showed similar knowledge to their matched controls without diabetes. Conclusions: T1D participants demonstrated the best carbohydrate knowledge compared to those with itT2D. Targeted educational interventions in itT2D populations may improve dietary management and clinical outcomes.

Abstract - NutriLens is an AI-powered nutrition tracking and recommendation system designed to improve dietary management through intelligent automation and personalization. The system enables users to log food using multimodal inputs such as image recognition, voice input, barcode scanning, and manual entry. It utilizes Generative AI and machine learning techniques to analyze food data and provide accurate nutritional insights, including calories, macronutrients, and micronutrients. A Retrieval-Augmented Generation (RAG) framework ensures reliable and context-aware responses by integrating verified nutritional knowledge. The platform also provides personalized recommendations and interactive dashboards for monitoring health metrics such as calorie intake, hydration, and weight trends. Developed using Python, Streamlit, and SQLite, NutriLens offers a scalable, user-friendly, and intelligent solution for modern nutrition management. Keyword - Artificial Intelligence, Machine Learning, Nutrition Tracking, Food Recognition, Retrieval-Augmented Generation (RAG), LSTM, Reinforcement Learning, Health Analytics, Multimodal AI.

ABSTRACT: With the rapid growth of artificial intelligence in healthcare and lifestyle management, automated dietary monitoring systems are becoming increasingly important. This project presents an intelligent food recognition and calorie measurement system using AI techniques. The system uses image processing and deep learning models to identify food items from images captured by a user through a mobile device or camera. A convolutional neural network (CNN) model is trained on a large dataset of food images to accurately classify different food categories. Once the food item is recognized, the system estimates its nutritional information, particularly calorie content, by referencing a pre-built food nutrition database. The proposed system aims to help users track their daily calorie intake automatically without manual logging, making diet monitoring more convenient and accurate. Such a system can support individuals who want to maintain a healthy lifestyle, manage weight, or monitor medical conditions like obesity and diabetes. Overall, this AI-based solution demonstrates how computer vision and machine learning can simplify dietary tracking and promote healthier eating habits. features from images. In a food recognition system, the CNN modelprocesses the captured food image and predicts the type

Accurate calorie and nutritional tracking is essential for maintaining healthy lifestyles, yet manual food logging remains time-consuming, error-prone, and abandoned by most users. This paper presents the Smart Nutrition Assistant v2.0, an AI-powered web application that uniquely combines state-of-the-art Computer Vision (CV) with deep learning techniques to automatically identify foods from photographs and estimate absolute portion sizes and nutritional values. The system employs Open-Vocabulary Food Recognition using CLIP (Contrastive Language-Image Pretraining) with the ViT-B-32 architecture, supporting over 100 food categories including Indian, Western, Asian, and Mexican cuisines without rigid categorical constraints. Advanced portion estimation leverages OpenCV-based plate detection through Hough Circle Transform, HSV-LAB color segmentation, and hemispherical volume approximation combined with food density mapping to derive accurate meal-level nutritional values scaled from per-100g standards. A multi-strategy barcode scanning pipeline integrates pyzbar, OpenCV BarcodeDetector, and OpenCV QRCodeDetector, querying the OpenFoodFacts API for packaged product nutrition. A pre-meal and post-meal comparative waste analysis module provides real-time breakdowns of calories consumed versus wasted. Experimental results demonstrate that the system achieves robust food recognition accuracy across diverse meal types with an intelligent fallback mechanism ensuring consistent performance even in resource-constrained environments. The proposed system offers a fast, reliable, and user-friendly solution for health-conscious individuals and patients managing dietary requirements.

Despite extensive development of flexible pressure sensors, it is still difficult for them to simultaneously achieve high precision and a large response to subtle pressures. To address these challenges, this work demonstrates a flexible pressure sensing platform that features the reduced graphene oxide aerogel sandwiched between a polydimethylsiloxane encapsulation layer and a thin polyimide film with interdigital electrodes. The resulting pressure sensor exhibits a high sensitivity of 698.96kPa-1 and a low limit of detection (~ 1Pa), and outstanding stability over 20,000 loading/unloading cycles. Besides monitoring various physiological signals and human motions, the flexible pressure sensors can be configured into an array layout as a smart artificial electronic skin to recognize the spatial pressure distribution. The flexible pressure sensor can also be integrated with signal processing and wireless communication modules as a teleoperation system for gesture recognition, force feedback control, and kitchen food recognition, highlighting future potential toward smart robotics and human-machine interfaces.

Reliable visual characterization of food composition is a fundamental prerequisite for image-based dietary assessment and health-oriented food analysis. In fine-grained food recognition, models often suffer from large intra-class variation and small inter-class differences, where visually similar dishes exhibit subtle yet discriminative differences in ingredient compositions, spatial distribution, and structural organization, which are closely associated with different nutritional characteristics and health relevance. Capturing such composition-related visual structures in a non-invasive manner remains challenging. In this work, we propose a fine-grained food classification framework that enhances spatial relation modeling and key-region awareness to improve discriminative feature representation. The proposed approach strengthens sensitivity to composition-related visual cues while effectively suppressing background interference. A lightweight multi-branch fusion strategy is further introduced for the stable integration of heterogeneous features. Moreover, to support reliable classification under large intra-class variation, a token-aware subcenter-based classification head is designed. The proposed framework is evaluated on the public FoodX-251 and UEC Food-256 datasets, achieving accuracies of 82.28% and 82.64%, respectively. Beyond benchmark performance, the framework is designed to support practical image-based dietary analysis under real-world dining conditions, where variations in appearance, viewpoint, and background are common. By enabling stable recognition of the same food category across diverse acquisition conditions and accurate discrimination among visually similar dishes with different ingredient compositions, the proposed approach provides reliable food characterization for dietary interpretation, thereby supporting practical dietary monitoring and health-oriented food analysis applications.

The rapid growth of health awareness and fitness tracking, accurate monitoring of daily food intake and calorie consumption has become increasingly important. Traditional methods of calorie tracking rely on manual data entry, which is time-consuming, error-prone, and inconvenient for users. To address these limitations, this project proposes a Food Recognition and Calorie Estimation System using Machine Learning, which automatically identifies food items from images and estimates their corresponding calorie values. The system uses computer vision and deep learning techniques to recognize different types of food from user-captured images. A convolutional neural network (CNN) model is trained on a labeled food image dataset to classify food items with high accuracy. Once a food item is recognized, its nutritional information, including calorie content, is retrieved from a predefined nutrition database. The system further estimates portion size using image-based features such as object area and volume approximation, enabling more accurate calorie calculation. The application provides a user-friendly interface where users can upload or capture food images, view recognized food names, and receive instant calorie estimates. This system is designed to support healthy lifestyle management by helping users track their daily calorie intake effortlessly. The proposed solution has potential applications in diet planning, fitness monitoring, and healthcare management. Experimental results demonstrate that the system achieves satisfactory recognition accuracy and reliable calorie estimation, making it a practical and efficient tool for real-world use.

A study on the generalization of DINOv2 features for food recognition tasks: A unified evaluation framework

The visual recognition of food depends on both domain-general and domain-specific visual mechanisms. Food-recognition ability negatively correlates with food neophobia (the tendency to avoid novel foods), possibly because the avoidance of novel food may limit perceptual experience that supports food recognition abilities. In prior work this relationship remained when domain-general object shape recognition ability (o) was controlled for, suggesting that this relationship is specific to food recognition. However, it is possible that other general recognition abilities, such as those for color and texture information, could also play a role. To test this, we developed outdoor scene-recognition tests. Like images of prepared food, outdoor scenes are rich in color and texture information. In 204 participants, we replicated previous findings that food-recognition tasks remain correlated after controlling for o. We additionally found that these correlations remain when scene-recognition ability is controlled for, strengthening evidence for a food-specific visual ability. The negative relationship between food-recognition ability and food neophobia also persisted after controlling for both o and scene-recognition ability, supporting the idea that this relationship reflects food-specific processes. These results indicate that food-specific processing goes beyond general visual abilities that apply to shape, color, and texture. The relationship with food neophobia supports a connection between perceptual expertise for food and affective responses to food that could inform interventions for individuals with restricted eating patterns.

Food waste remains a global challenge, particularly among younger generations. This study examines the attitudes and behaviours of 330 Generation Z individuals (aged 18-24 years) from Italy, Estonia, Croatia, Romania, and Serbia using an extended Theory of Planned Behaviour (TPB). The TPB model was expanded to include moral social values, awareness of health risks, and good provider identity. A mixed-methods approach was applied, combining 7-day food waste diaries, visual plate-waste analysis, and self-administered questionnaires. Food recognition analysis showed that Estonian participants wasted less food per meal (3.43%) than those from Italy, Serbia, Croatia, and Romania (12.53%, 12.57%, 14.53%, and 17.18%). Nationality-specific patterns emerged: Romanians mainly discarded meat and potatoes, while participants from Estonia, Croatia, and Serbia wasted fruit and vegetables; Italians most frequently wasted fish and dairy. The extended TPB effectively predicted intentions to reduce food waste, identifying key behavioural determinants that can inform targeted interventions for young consumers.

In modern diets, increasing consumption of ultra-processed and high-calorie foods is accelerating obesity and metabolic disorders, highlighting the need for technologies that automatically record and manage food types and nutritional intake. This study proposes a mobile application architecture that integrates a Swin-V2-based deep learning model for food image classification with a basal metabolic rate (BMR)-based meal recommendation module. Experiments use 10,000 images from 10 frequently consumed classes selected from the 101-class Food-101 dataset, with the limitation that computational and time constraints prevented evaluation on all classes. The proposed Swin-V2-Tiny model, fine-tuned from ImageNet pre-trained weights, achieves a Top-1 accuracy of 0.926 and a macro F1-Score of 0.926, outperforming Swin-V1 and several CNN-based baselines. The prototype system shows that automatic food recognition and diet recommendation can be effectively combined, and future work will extend the dataset, incorporate portion-size estimation, and apply model compression to build an integrated diet-management system suitable for real-world services.

Deep learning-based food recognition has made significant progress in predicting food types from eating occasion images. However, two key challenges hinder real-world deployment: (1) continuously learning new food classes without forgetting previously learned ones, and (2) handling the long-tailed distribution of food images, where a few common classes and many more rare classes. To address these, food recognition methods should focus on long-tailed continual learning. In this work, We introduce a dataset that encompasses 186 American foods along with comprehensive annotations. We also introduce three new benchmark datasets, VFN186-LT, VFN186-INSULIN and VFN186-T2D, which reflect real-world food consumption for healthy populations, insulin takers and individuals with type 2 diabetes without taking insulin. We propose a novel end-to-end framework that improves the generalization ability for instance-rare food classes using a knowledge distillation-based predictor to avoid misalignment of representation during continual learning. Additionally, we introduce an augmentation technique by integrating class-activation-map (CAM) and CutMix to improve generalization on instance-rare food classes. Our method, evaluated on Food101-LT, VFN-LT, VFN186-LT, VFN186-INSULIN, and VFN186-T2DM, shows significant improvements over existing methods. An ablation study highlights further performance enhancements, demonstrating its potential for real-world food recognition applications.

<p indent="0mm">Nutrition informatics has undergone a significant paradigm shift in recent years. Approaches historically grounded in rule-based decision support and classical task-specific machine learning pipelines are increasingly being superseded by an ecosystem centered on large language models (LLMs) and multimodal vision-language foundation models. This review synthesizes researches published between 2019 <?A3B2 pi129?>and 2025, with the objectives of clarifying architectural patterns that enable nutrition-oriented perception and reasoning, summarizing advances and identifying gaps across major application scenarios, and outlining strategic directions for reliable translation research in clinical and public health practice. Based on a systematic analysis of 92 <?A3B2 pi129?>representative studies, we organize the current landscape into three interrelated research trajectories: (1) Vision and multimodal modeling for dietary perception, focusing on food recognition, ingredient parsing, portion estimation, and nutrient prediction from meal images and videos. Recent methodologies increasingly adopt Transformer-based encoders and explicit vision-language alignment, leveraging depth cues and scale calibration to improve robustness under complex real-world conditions. (2) LLM-based nutrition agents for interactive guidance, supporting dietary counseling, meal planning, and health coaching. To mitigate challenges such as hallucinations and numerical inconsistency, current research emphasizes domain adaptation, tool-augmented computation, and retrieval-augmented generation (RAG) to ground model responses in reliable nutrition databases and clinical guidelines. (3) Personalization-oriented hybrid systems, which combine foundation models with structured components—such as knowledge graphs and causal inference frameworks—while integrating individual-level multi-omics signals, biomarkers, and lifestyle data. These systems aim to generate and optimize meal plans under strict constraints of safety, clinical feasibility, and patient adherence. Across these trajectories, interpretability has transitioned from an optional feature to a core system requirement, driven by the needs of clinical accountability and risk auditing. Concurrently, evaluation protocols are expanding from image-centric datasets (e.g., Nutrition5k) to comprehensive<styleredit/> benchmarking suites designed for multimodal reasoning. Despite rapid progress, limitations persist regarding model factuality, privacy preservation, and external validity across diverse cuisines and socioeconomic settings. We advocate for evidence-grounded pipelines, standardized multimodal datasets with clinical endpoints, and unified evaluation frameworks spanning accuracy, safety, and bias. Human-in-the-loop deployment remains essential to quantify benefit-risk profiles and facilitate the regulatory adoption of AI-driven nutrition services. <alternatives id="alt2"> <graphic xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="./online/c2.png" href="./online/c2.png" specific-use="online"/> <graphic href="./print/c2.pdf" xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="./print/c2.pdf" specific-use="print"/> </alternatives>

Accurately monitoring diet remains a challenge, especially for home-cooked or restaurant meals that do not carry nutritional labels and are difficult to record using conventional calorie-counting apps. Most existing tools rely on manual entry or barcode scanning, which are time-consuming and often inaccurate. To address these limitations, this study introduces an AI-powered calorie tracking system that combines image captioning (BLIP model) with large language models (Groq LLM via LangChain) to automatically recognize food items, estimate their nutritional values, and generate personalized feedback. Users can simply upload a photo of a meal or enter food names, and the system provides calorie and nutrient breakdowns, daily summaries, and fitness suggestions when caloric excess is detected. Reports are automatically compiled into a downloadable PDF for long-term monitoring. The system demonstrated strong performance, achieving a precision of 92.4%, a recall of 91.1%, and an F1-score of 91.7% in food recognition, with an AUC of 0.95 for nutritional estimation. Case studies confirmed its ability to handle multi-item meals, correctly identify key nutrients, and provide contextual health advice. To ensure transparency and trust, the framework incorporates Explainable AI (XAI) techniques such as Grad-CAM, SHAP, LIME, and EDGE, which help users and healthcare professionals understand how predictions are generated. By reducing manual effort, adapting to food variability, and offering real-time, interpretable insights, this system provides a practical, reliable, and user-friendly solution for personal health management, chronic disease care, and commercial fitness applications.

Artificial intelligence (AI) is progressively transforming the field of nutrition, evolving from early applications focused on food recognition, dietary intake estimation, and malnutrition screening to more complex systems capable of integrating clinical, anthropometric, biochemical, dietary, functional, and contextual data. Within this emerging landscape, artificial intelligence agents, or agentic AI systems, represent a new generation of digital tools characterized by their ability to reason, plan, use external tools, maintain longitudinal memory, and coordinate tasks within predefined boundaries. This review analyzes the potential role of AI agents in nutrition based on some recent studies addressing their application in digital health, public nutrition, nutritional oncology, laboratory medicine, and personalized preventive programs. The available evidence suggests that AI agents may provide value in three major areas: automated and multimodal nutritional assessment, generation of personalized dietary recommendations, and longitudinal coordination of nutritional interventions in patients with chronic diseases, cancer, or cardiometabolic risk. In public health, agentic models may facilitate scalable precision nutrition strategies, particularly in communities with limited access to dietitians, nutrition specialists, or healthcare resources. In oncology, these systems could integrate malnutrition screening, body composition assessment, symptoms, oncological treatments, and dietary preferences in order to anticipate nutritional deterioration and activate early interventions. In laboratory medicine, AI agents may link biomarkers, omics data, and clinical outcomes to support more dynamic and individualized nutritional decision-making. Their implementation poses important challenges, including insufficient clinical validation, risk of errors or hallucinations, cultural and socioeconomic biases, privacy concerns, professional responsibility, interoperability, and the need for continuous human supervision. In conclusion, agentic AI represents a relevant evolution for both clinical and community nutrition. Nevertheless, its incorporation into practice should be gradual, regulated, explainable, prospectively evaluated, and subordinated to clinical judgment and patient values.

Accurate food recognition and nutrient estimation are critical for practical dietary assessment, health monitoring, and personalized nutrition management. Traditional calorie tracking methods rely heavily on manual input and self-reporting, which are often inaccurate, time-consuming, and inconsistent. Existing AI-based food recognition systems primarily rely on basic convolutional neural networks and two-dimensional image analysis, limiting their ability to identify complex, mixed, and regional dishes and failing to estimate portion sizes accurately. These limitations significantly reduce their practical applicability, particularly for diverse cuisines such as Indian food. To address these challenges, this project presents Nutri Vision, an AI-driven framework for food recognition, portion size estimation, nutrient prediction, and personalized dietary guidance. The proposed system integrates advanced deep learning models including YOLOv8, Vision Transformers, and Region-Based Convolutional Neural Networks to accurately detect and classify multiple food items from a single image. Portion size estimation is achieved using pixel-to-gram conversion and depth-aware analysis, enabling reliable calorie and nutrient computation through integrated food databases. Furthermore, machine-learning-based decision models are employed to generate personalized diet recommendations and healthier food alternatives based on users’ goals and health conditions. The system delivers real-time, culturally adaptive, and scalable nutrition insights with high accuracy, making it suitable for applications in healthcare, fitness management, and nutrition research.

The rapid evolution of artificial intelligence (AI) and deep learning has transformed the field of nutritional analysis, offering significant improvements over traditional methods in food recognition and calorie estimation. Conventional techniques based on convolutional neural networks (CNNs) have shown promise yet remain limited by extensive data requirements, language dependence, and inadequate nutritional insights. In this paper, we propose a novel, multilingual, generative AI-based approach that leverages large multimodal models (LMMs) such as Google’s Gemini Pro Vision and OpenAI’s GPT-4 Vision. Our solution integrates robust image validation, dynamic prompt engineering, and multilingual natural language processing to deliver detailed calorie estimates and nutritional breakdowns while overcoming the challenges inherent in CNN-based systems. We detail the underlying algorithms, provide a conceptual system flowchart, and present comparative analyses against traditional approaches. Finally, our consolidated “Proposed Solution and Future Directions” section describes the system architecture, implementation details, and outlines the future research agenda.

The rising global burden of chronic diseases highlights the limitations of traditional dietary guidelines. Precision Nutrition (PN) aims to deliver personalized dietary advice to optimize individual health, and the effective implementation of PN fundamentally relies on comprehensive and accurate dietary data. However, conventional dietary assessment methods often suffer from quantification errors and poor adaptability to dynamic changes, leading to inaccurate data and ineffective guidance. Machine learning (ML) offers a powerful suite of tools to address these limitations, enabling a paradigm shift across the nutritional management pipeline. Using dietary data as a thematic thread, this article outlines this transformation and synthesizes recent advances across dietary assessment, in-depth mining, and nutritional intervention. Additionally, current challenges and future trends in this domain are also further discussed. ML is driving a critical shift from a subjective, static mode to an objective, dynamic, and personalized paradigm, enabling a loop nutrition management framework. Precise food recognition and nutrient estimation can be implemented automatically with ML techniques like computer vision (CV) and natural language processing (NLP). Integrating with multiple data sources, ML is conducive to uncovering dietary patterns, assessing nutritional status, and deciphering intricate nutritional mechanisms. It also facilitates the development of personalized dietary intervention strategies tailored to individual needs, while enabling adaptive optimization based on users' feedback and intervention effectiveness. Although challenges regarding data privacy and model interpretability persist, ML undeniably constitutes the vital technical support for advancing PN into practical reality.