EXPERIMENTAL STUDY

Acurateţea inteligenţei artificiale în analiza imaginilor cu alimente

 Accuracy of artificial intelligence in food image analysis

First published: 28 iunie 2024

Editorial Group: MEDICHUB MEDIA

DOI: 10.26416/JourNutri.2.2.2024.9766

Abstract

Assessing the nutritional values of meals is important for the public health. Current evaluation methods can be imprecise due to human errors, but food image ana­ly­sis, facilitated by artificial intelligence (AI), can en­hance determination accuracy. This study examines the accuracy of GPT-4 in identifying foods and estimating nu­tri­tio­nal values from simple images. Methodology. Three images of culinary dishes were used. A customized GPT model was engaged to identify the foods and esti­mate their nutritional values. The results were compared with those from the Nutrition ARTS platform, which uses a USDA database. Results. The GPT model correctly iden­ti­fied the foods in all images. The estimated calorie va­lues had an average difference of ±5 kcal compared to the reference. Differences for proteins were ±4.4 g, for car­bo­hy­drates ±1.3 g, and for fats ±0.7 g. Differences in micronutrient values were higher. Conclusions. The GPT model accurately identified foods and estimated ca­lo­ries and macronutrients. However, differences in micronutrient values were significant. The study high­lighted the potential and limitations of AI in food image ana­ly­sis. Further research with a larger dataset could pro­vide more relevant information.
 

Keywords
nutritional analysis, food images, artificial intelligence, nutritional values

Rezumat

Determinarea valorilor nutriţionale ale meselor este im­por­tan­tă pentru sănătatea publică. Metodele actuale de evaluare pot fi imprecise din cauza erorilor umane, dar analiza ima­gi­ni­lor cu alimente, facilitată de inteligenţa artificială (IA), poate spo­ri precizia determinărilor. Studiul de faţă examinează acu­ra­te­ţea GPT-4 în identificarea alimentelor şi estimarea va­lo­ri­lor nu­tri­ţio­nale din imagini simple. Metodologie. S-au utilizat trei imagini cu preparate culinare. Un model per­so­na­li­zat GPT a fost folosit pentru a identifica alimentele şi a estima va­lo­ri­le nutriţionale. Rezultatele au fost comparate cu cele ale platformei Nutrition ARTS, care foloseşte o bază de date a USDA. Rezultate. Modelul GPT a identificat corect alimentele din toate imaginile. Valorile caloriilor estimate au avut o di­fe­ren­ţă medie de ±5 kcal faţă de referinţă. Diferenţele pen­tru proteine au fost de ±4,4 g, pentru glucide de ±1,3 g şi pen­tru lipide de ±0,7 g. La micronutrienţi, diferenţele au fost mai crescute. Concluzii. Modelul GPT a demonstrat o bună acu­ra­te­ţe în identificarea alimentelor şi estimarea caloriilor şi macronutrienţilor. Totuşi, pentru micronutrienţi, diferenţele au fost semnificative. Studiul a evidenţiat potenţialul, dar şi limitările IA în analiza imaginilor cu alimente. Cercetări ul­te­rioa­re, pe un set mai mare de date, vor putea oferi informaţii mai relevante.
 

Bibliografie

  1. Konstantakopoulos FS, Georga EI, Fotiadis DI. A Review of Image-Based Food Recognition and Volume Estimation Artificial Intelligence Systems. IEEE Rev Biomed Eng. 2024;17:136-152.

  2. Cena H, Calder PC. Defining a healthy diet: evidence for the role of contemporary dietary patterns in health and disease. Nutrients. 2020;12(2):334.

  3. Shonkoff E, Cara KC, Pei XA, Chung M, Kamath S, Panetta K, Hennessy E. AI-based digital image dietary assessment methods compared to humans and ground truth: a systematic review. Ann Med. 2023;PP(99).

  4. Gowda S. Improving Multi-Modal Food Detection System with Transfer Learning. LMU/LLS Theses and Dissertations. 2023;1238.

  5. Min W, Wang Z, Liu Y, et al. Large-scale visual food recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2023; arXiv:2103.16107. 

  6. Sak J, Suchodolska M. Artificial intelligence in nutrients science research: a review. Nutrients. 2021;13(2):322.

  7. Darwish A, Ricci M, Zidane F, et al. Physical contamination detection in the Food Industry using microwave and machine learning. Electronics. 2022;11(19):3115.

  8. Chhikara P, Chaurasia D, Jiang Y, et al. FIRE: Food Image to REcipe Generation. 2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). 2024;8169-8179.

  9. King RC, Bharani V, Shah K, et al. GPT-4V passes the BLS and ACLS examinations: An analysis of GPT-4V’s image recognition capabilities. Resuscitation. 2024;195:110106.

  10. Kagaya H, Aizawa K, Ogawa M. Food detection and recognition using convolutional neural network. In Proceedings of the 22nd ACM International Conference on Multimedia. 2014;1085-1088.

  11. Johnson OV, Alyasiri OM, Akhtom D, Johnson OE. Image Analysis through the lens of ChatGPT-4. Journal of Applied Artificial Intelligence. 2023;4(2).

  12. Mengsuwan K, Palacio JCR, Ryo M. ChatGPT and general-purpose AI count fruits in pictures surprisingly well. arXiv preprint. 2024; arXiv:2404.08515.

  13. AlZu’bi S, Mughaid A, Quiam F, Hendawi S. Exploring the capabilities and limitations of chatgpt and alternative big language models. Artificial Intelligence and Applications. 2024;2(1):28-37.

  14. Ponrawin K, Pongpipat P, Natthanet T. Smart Cuisine: Generative recipe & ChatGPT powered nutrition assistance for sustainable cooking. Procedia Computer Science. 2023;225:2028-2036.

  15. Côté M, Lamarche B. Artificial intelligence in nutrition research: perspectives on current and future applications. Applied Physiology, Nutrition, and Metabolism. 2022;47(1):1-8.

  16. Limketkai BN, Mauldin K, Manitius N, Jalilian L, Salonen BR. The age of artificial intelligence: use of digital technology in clinical nutrition. Current Surgery Reports. 2021;9(7):20.

  17. Vrapcea G, Tarcea M. The importance of nutrition care platforms for Romanian dietitians’ practice. Health, Sports & Rehabilitation Medicine. 2022;23(3):114-119.