An Evaluation of ChatGPT for Nutrient Content Estimation from Meal Photographs

Nutrients, Год журнала: 2025, Номер 17(4), С. 607 - 607

Опубликована: Фев. 7, 2025

Background/Objectives: Advances in artificial intelligence now allow combined use of large language and vision models; however, there has been limited evaluation their potential dietary assessment. This study aimed to evaluate the accuracy ChatGPT-4 estimating nutritional content commonly consumed meals using meal photographs derived from national survey data. Methods: Meal (n = 114) were uploaded ChatGPT it was asked identify foods each meal, estimate weight, nutrient for 16 nutrients comparison with known values precision, paired t-tests, Wilcoxon signed rank test, percentage difference, Spearman correlation (rs). Seven dietitians also estimated energy, protein, carbohydrate thirty-eight intraclass (ICC). Results: Comparing actual meals, showed good precision (93.0%) correctly identifying photographs. There agreement weight (p 0.221) small but poor medium < 0.001) meals. 10 0.05). Percentage difference >10% 13 nutrients, underestimating 11 nutrients. Correlations adequate or all rs ranging 0.29 0.83. When comparing dietitians, ICC ranged 0.31 0.67 across Conclusions: performed well foods, weights portion sizes, ranking according content, poorly sizes providing accurate estimates content.

Язык: Английский

---

Self-perceived food literacy in relation to the quality of overall diet and main meals: A cross-sectional study in Japanese adults

Appetite, Год журнала: 2024, Номер 196, С. 107281 - 107281

Опубликована: Фев. 17, 2024

This cross-sectional study aimed to assess the relationship between self-perceived food literacy (SPFL) and quality of overall diet main meals in Japanese adults. In total, 5998 adults aged 20–79 years were included this analysis. The SPFL was assessed using version 29-item Dutch scale (score range 1–5). Using validated dietary information, Healthy Eating Index-2015 (HEI-2015) calculated 0–100). mean 3.18; internal consistency considered good (Cronbach's alpha: 0.80). HEI-2015 for 50.4. significantly positively associated with HEI-2015. multiple linear regression, one point increase corresponded an by a 4.8 diet, 6.2 breakfast, 4.6 lunch, 3.6 dinner (all P < 0.0001). Six eight domains (i.e., preparation skills, resilience resistance, healthy snack styles, examining labels, budgeting, stockpiling) diet. When each meal examined, showing significant associations all three budgeting. stockpiling breakfast but not dinner. social conscious eating daily planning only, resistance only. conclusion, cross-sectionally Further observation intervention studies are needed confirm observed here.

Язык: Английский

---

Benchmarking nutrition facts panel label – a consumer ethics perspective using health belief model

Benchmarking An International Journal, Год журнала: 2024, Номер unknown

Опубликована: Апрель 26, 2024

Purpose This study examines how health-conscious consumers utilize nutrition facts panel labels when purchasing food products, focusing specifically on the dimension of ethical evaluation. It aims to understand considerations influence decision-making process who prioritize health. By analyzing impact evaluation label usage, sheds light significance ethics in consumer behavior context packaged edible oil. Design/methodology/approach Empirical data were collected using an online survey and a non-ordered questionnaire. In total, 469 valid responses obtained. The used SPSS version 27.0 SmartPLS 3 for demographic analysis structural equation modeling. Findings findings suggest that three factors – perceived benefits, threats, self-efficacy, positively use NFP labels. However, barriers negatively additionally, mediates usage Practical implications health belief model, functions as mediator has greater use. provides framework marketers promote consciousness by encouraging them incorporate Originality/value is one first attempts demonstrate mediate beliefs

Язык: Английский

---

Content accuracy and readability of dietary advice available on webpages: A systematic review of the evidence

Journal of Human Nutrition and Dietetics, Год журнала: 2024, Номер 38(1)

Опубликована: Ноя. 26, 2024

Abstract Background Concerns have been raised regarding the quality of health information published on World Wide Web, while studies accumulating similar evidence for nutrition‐related are scarce. The present review aimed to systematically accumulate and discuss findings evaluating content websites publishing information, based PRISMA statement. Methods Studies that assessed accuracy readability dietary advice websites/webpages were evaluated. SPIDER framework was used systematic search studies, those evaluated did not contain nutritional videos or social media, referred professionals, contained a specific theme (e.g., supplements), excluded. Results Twenty‐nine included in this review, assessing mainly English, 18 different diet‐related topics. Twenty them classified reported, also, results per category. Inconsistent has found 16%–49.6% websites, with percentage rising 54%–94% case ‘weight loss’ information. Purely congruent guidelines 18%–39.7% websites. Commercial sites inferior terms rest. level estimated as higher than recommended 9 out 11 it. Conclusions indicate inaccurate hard‐to‐read is many regardless topic.

Язык: Английский

---

Reevaluating feature importances in machine learning models for schizophrenia and bipolar disorder: The need for true associations

Brain Behavior and Immunity, Год журнала: 2024, Номер 124, С. 123 - 124

Опубликована: Ноя. 29, 2024

Язык: Английский

---