A systematic review of AI-based food and calorie detection systems in healthcare: Trends, techniques, and applications

Abstract

Background: Accurate assessment of calorie and nutrient intake is essential for the prevention and management of chronic diseases such as obesity and type 2 diabetes. However, traditional dietary assessment methods, including food diaries and dietary recalls, are often labor-intensive, time-consuming, and prone to reporting errors. Objective: This review aims to systematically examine the current landscape of artificial intelligence (AI)-based food recognition and calorie estimation systems in healthcare, highlighting technological developments, applications, performance trends, and research gaps. Methods: A systematic review was conducted following PRISMA 2020 guidelines. Literature published between 2015 and 2025 was retrieved from Scopus, PubMed, IEEE Xplore, Web of Science. Google Scholar was additionally searched to identify relevant grey literature and citation records. Studies focusing on AI-driven dietary assessment using computer vision, machine learning, multimodal learning, and large language models (LLMs) were screened according to predefined inclusion and exclusion criteria. Results: A total of 57 studies met the eligibility criteria. Three major technological trends were identified: (1) convolutional neural network (CNN)-based approaches, achieving 80–97% classification accuracy; (2) transformer-based architectures, including Vision Transformers and Swin Transformers reported accuracies ranging up to 99.83% under specific benchmark settings; and (3) multimodal and LLM-based systems, such as GPT-4V, Gemini, and Claude, demonstrating 89.8% food recognition accuracy and promising portion estimation performance. Commercial applications including MyFitnessPal, CalorieMama, DietCam, and FoodAI were also reviewed. Conclusion: AI-driven food recognition and dietary assessment technologies have significantly advanced nutritional monitoring and personalized healthcare. Nevertheless, challenges remain in portion estimation, clinical validation, and the representation of culturally diverse cuisines, particularly in South Asian contexts. Future research should focus on region-specific datasets, multimodal AI systems, and clinically validated deployment frameworks.