資料載入中.....
|
請使用永久網址來引用或連結此文件:
http://libir.tmu.edu.tw/handle/987654321/64618
|
| 題名: | Digital Nutrition: The Development, Integration, and Evaluation of Technology-Based Dietary Assessment in Dietetic Curriculum |
| 作者: | NGAN, HO DANG KHANH |
| 貢獻者: | 保健營養學系博士班
張榮素 |
| 關鍵詞: | dietary assessment;mhealth;nutrition app |
| 日期: | 2024-06-12 |
| 上傳時間: | 2024-11-06 14:58:01 (UTC+8) |
| 摘要: | 營養評估領域正在迅速發展，採用了數位成像、移動和網路應用程式、擴增實境（AR）和人工智慧（AI）等技術。然而，這些技術與金標準的有效性對比仍未得到充分探討，且正式的營養學培訓課程尚未完全整合這些進展。本論文通過五項研究解決了這一差距，主要有兩個目標：
(1) 開發、整合和評估兩個創新工具在營養學培訓中的應用：使用3D模型在AR環境中的互動份量教育平台（3D AR增強）（研究1和2）和Formosa FoodApp，一個多語言的圖像輔助營養應用程式（研究3和4）。參與者是2019年至2023年間台灣臺北醫學大學的營養學學生。
(2) 進行一項網絡統合分析（研究5），評估數位營養評估工具相對於傳統和基於生物標誌物的方法在能量和營養攝入評估中的準確性。
具體來說：
研究1結合了一個32週的框架，將2D和3D食品模型整合到營養學培訓中，共有65名未來的營養師參與。結果顯示，在食品識別和量化技能方面，2D和3D模型之間沒有顯著差異（食品識別：2D: 89% vs. 3D: 85%；量化：2D: 19.4% vs. 3D: 19.3%），但3D模型在食品量化技能方面可能有所幫助。重複訓練顯著提高了食品識別和量化技能，甚至對那些初期表現較差的學生也是如此。
研究2引入了一個3D AR增強平台，以改善份量估算和學習體驗。67名學生參與了研究，結果顯示3D-AR模型在量化準確性方面優於真實食品視覺估算（RFVE）（27% vs. 19%，p<0.001）。經過32週的訓練後，食品量化技能顯著提高（從27%到40%），學生表示更喜歡3D模型，因為其穩定性和易用性。
研究3開發並評估了一個多語言的圖像輔助營養應用程式——Formosa FoodApp，與24小時飲食回顧（24-HDR）對比，以解決移動營養評估中的測量誤差。手動數據清理發現12.4%的食品編碼錯誤和32%的選定食品編碼缺失微量營養素。常見的即食預包裝食品、超加工食品及餐館/街頭食品缺乏微量營養素數據，是導致該移動營養評估方法中最大測量誤差的原因。重新分析食品編碼提高了微量營養素攝入的準確性，增強了應用程式與24-HDR之間的相關性。
研究4評估了Formosa FoodApp和一些商業營養應用程式（COFIT、MyFitnessPal-中文版、MyFitnessPal-英文版和LoseIt!）在追蹤飽和脂肪和膽固醇攝入方面的可靠性。與USDA-FNDDS和台灣食品成分資料庫（TaiwanFCD）對比，所有商業應用程式在飽和脂肪（-13.8%到-40.3%）和膽固醇（-26.3%到-60.3%）方面均發現顯著錯誤。高變異性和數據遺漏情況顯著，MyFitnessPal在不同國家食品成分資料庫中表現出一致的錯誤。
總之，本論文強調了數位技術在增強營養學培訓和營養評估準確性方面的潛力，同時也指出了現有的挑戰和不準確性。特別是在不同環境和年齡組中，迫切需要提高數位營養評估工具的準確性，尤其是基於圖像和AI的工具。
The dietary assessment (DA) field is rapidly evolving with technologies like digital imaging, mobile and web applications, Augmented Reality (AR), and Artificial Intelligence (AI). However, their validity against gold standards remains underexplored, and formal dietetic training programs have not fully integrated these advancements. The dissertation addressed this gap through five studies with two primary objectives:
(1) To develop, integrate, and evaluate two innovative tools in dietetic training: the Interactive Portion Size Education platform using 3D models in AR environment (3D AR-enhanced) (Studies 1 and 2) and the Formosa FoodApp, a multilingual, image-assisted nutrition app (Studies 3 and 4). Participants were dietetic students from Taipei Medical University, Taiwan, from 2019 to 2023.
(2) To conduct a network meta-analysis (Study 5) evaluating the accuracy of digital DA tools against traditional and biomarker-based methods for assessing energy and nutrient intake.
Specifically:
Study 1 integrated a 32-week framework combining 2D and 3D food models in dietetic training among 65 future dietitians. No significant differences in food identification and quantification skills between 2D and 3D models was observed (food identification: 2D: 89% vs. 3D: 85%; quantification: 2D: 19.4% vs. 3D: 19.3%), with 3D models potentially aiding food quantification skill. Repeated training significantly improved food identification and quantification skills even for students who initially struggled.
Study 2 introduced a 3D AR-enhanced platform to improve portion size estimation and learning experience. Among 67 students, 3D-AR models outperformed real food visual estimation (RFVE) in quantification accuracy (27% vs. 19%, p<0.001). Food quantification skills improved significantly after 32 weeks (from 27% to 40%), with students expressing a preference for 3D models due to stability and ease of use.
Study 3 developed and evaluated a multilingual, image-assisted nutrition app - Formosa FoodApp against the 24-hour dietary recall (24-HDR) to address measurement errors in mobile-based DAs. Manual data cleaning identified 12.4% food coding errors and 32% selected food codes with missing micronutrients were re-analyzed. The lack of micronutrient data for commonly consumed ready-to-eat pre-packaged and ultra-processed foods as well as restaurant/street foods explains the greatest measurement errors in this mobile-based DA method. Reanalyzing food codes improved the accuracy of micronutrient intake and enhanced correlations between the app and 24-HDR.
Study 4 evaluated Formosa FoodApp and commercial nutrition apps (COFIT, MyFitnessPal-Chinese, MyFitnessPal-English, and LoseIt!) for reliability in tracking saturated fat and cholesterol intake. Compared to USDA-FNDDS and TaiwanFCD databases, significant errors were found in saturated fat (-13.8% to -40.3%) and cholesterol (-26.3% to -60.3%) across all commercial apps. High variability and data omissions were noted, with MyFitnessPal showing consistent errors across different national food composition databases (FCDs).
In summary, the dissertation emphasized the potential of digital techonology in enhancing dietetic training and DA accuracy, though they also highlight existing challenges and inaccuracies. There is a strong need for improved accuracy in digital DA tools, particularly focusing on image-based and AI tools across different settings and among different age groups. |
| 描述: | 博士
指導教授：張榮素
口試委員：趙振瑞
口試委員：姚智原
口試委員：邱雪婷
口試委員：潘文涵
口試委員：張榮素 |
| 附註: | 論文公開日期：2024-06-21 |
| 資料類型: | thesis |
| 顯示於類別: | [保健營養學系暨研究所] 博碩士論文
|
文件中的檔案:
| 檔案 |
描述 |
大小 | 格式 | 瀏覽次數 |
|---|
| index.html | | 0Kb | HTML | 67 | 檢視/開啟 |
|
在TMUIR中所有的資料項目都受到原著作權保護.
|
