AC/DC: Artificial intelligence and Computer visioning to assess Dietary Composition

AC/DC：人工智能和计算机视觉评估膳食成分

• 批准号: 9978495
• 负责人: Erin Hennessy
• 金额: $ 22.2万
• 依托单位: TUFTS UNIVERSITY BOSTON
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-05-13 至 2022-04-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9978495
• 关键词: 3-Dimensional; Address; Algorithms; Artificial Intelligence; Assessment tool; Behavior; Beverages; Body Weight decreased; Calories; Cause of Death; Cellular Phone; Code; Complex; Computer Vision Systems; Consumption; Data; Databases; Detection; Development; Diet; Diet Records; Dietary Assessment; Dietary Intervention; Dietary Practices; Dietary intake; Economics; Engineering; Epidemic; Food; Goals; Gold; Health; Healthcare; Home environment; Human; Human Resources; Hybrids; Image; Imaging Techniques; Individual; Intake; Intervention; Knowledge; Left; Link; Measurement; Measures; Methodology; Methods; Monitor; Nutrient; Nutritional Science; Nutritional status; Obesity; Obesity Epidemic; Outcome; Output; Participant; Research; Research Personnel; Research Training; Respondent; Restaurants; Side; Speed; System; Techniques; Technology; Testing; Time; Training; Validation; Weight; Work; base; computer science; cost; cost effective; design; digital; digital imaging; disorder risk; expectation; food consumption; food quality; handheld mobile device; improved; knowledge base; new technology; novel; nutrition; prototype; success; system architecture; tool; virtual model; wasting; weight maintenance

ABSTRACT Dietary intake is a complex human behavior that drives disease risk and corresponding economic and healthcare burdens worldwide. Poor diet is the leading cause of death in the US and a known driver of obesity – a global epidemic. A major contributor to poor diet is food eaten away from home, such as restaurant foods. Research has shown that tracking one’s weight and dietary intake significantly improve success toward weight loss and maintenance goals; however, this type of tracking is burdensome, prone to error, and difficult to estimate for restaurant foods. Accurate approaches and tools to evaluate food and nutrient intake are essential in monitoring the nutritional status of individuals. There is a critical need for real-time data capture that minimizes burden and reduces error. While progress has been made, there is no tool available that accurately and automatically estimates foods left unconsumed in a meal. Two major limitations of existing systems is the reliance of a fiducial marker for food detection and volume estimation, and reliance on humans – either the respondent or a trained researcher – to estimate the portion of food leftover. This application leverages novel technology to remove those limitations. The long-term research goal is to utilize digital imaging (DI), artificial intelligence (AI) and computer vision (CV) techniques to develop a novel hybrid methodology for rapid, accurate measurement of dietary intake. To attain this goal, our objective in this R21 application is to refine and test a system architecture that (a) uses digital images to record dietary intake in real-time and (b) uses AI and CV techniques to identify food/beverage items and determine amounts leftover. We plan to build on our current prototype in which digital food images are captured before and after the meal, analyzed to detect the food items, a three-dimensional (3-D) virtual model constructed, and volume remaining after the meal estimated, which will be used to calculate the amount leftover based on the initial volume. Volume consumed will be converted to weight and linked to public-use nutrition information. These calorie estimates will be compared against calories those from (a) DIs coded by trained research staff and (b) weighed plate waste methodology. Our expectation is to develop a valid system architecture for rapidly estimating dietary intake. The outcome of this proposal is expected to have a significant positive impact, enabling nutrition and health researchers to collect high-quality food consumption data in real world settings, increasing knowledge of dietary patterns and improving capacity to assess dietary interventions. This work will lead to an R01 application that will expand food types and meal settings and test the utility of our system among consumers.

摘要 饮食摄入是一种复杂的人类行为，它会导致疾病风险和相应的经济和社会风险。 全世界的医疗负担。不良的饮食习惯是美国人死亡的主要原因，也是众所周知的肥胖症的驱动因素 - 一种全球性的流行病导致饮食不良的一个主要原因是在家里吃的食物，例如餐馆的食物。 研究表明，跟踪一个人的体重和饮食摄入量显着提高成功的体重 损失和维护目标;然而，这种类型的跟踪是繁重的，容易出错，并且难以 估计餐厅的食物。评估食物和营养摄入的准确方法和工具至关重要 监测个人的营养状况。对于实时数据捕获的迫切需要， 最大限度地减少负担并减少错误。虽然已经取得了进展，但没有任何工具可以准确地 并自动估计一顿饭中未食用的食物。现有系统的两个主要局限性是 食物检测和体积估计的基准标记的依赖，以及对人类的依赖- 受访者或训练有素的研究人员-估计剩余食物的比例。该应用程序利用了新颖的 技术来消除这些限制。长期的研究目标是利用数字成像（DI），人工 智能（AI）和计算机视觉（CV）技术， 准确测量饮食摄入量。为了实现这一目标，我们在R21应用程序中的目标是改进 并测试一个系统架构，该架构（a）使用数字图像实时记录饮食摄入量，（B）使用人工智能 和CV技术来识别食品/饮料项目并确定剩余量。我们计划建立在我们的 当前的原型，其中在餐前和餐后捕获数字食物图像，分析以检测 食物项目、构建的三维（3-D）虚拟模型以及餐后剩余的体积 估计，这将用于计算基于初始体积的剩余量。消耗量 将被转换为体重，并与公共营养信息相关联。这些卡路里估计将是 与来自（a）经培训的研究人员编码的DI和（B）称重的平板废物的卡路里进行比较 方法论我们的期望是开发一个有效的系统架构，快速估计膳食摄入量。 预计这一提案的结果将产生重大的积极影响， 研究人员在真实的世界环境中收集高质量的食物消费数据， 饮食模式和提高评估饮食干预措施的能力。这项工作将导致R 01 该应用程序将扩展食品类型和膳食设置，并测试我们的系统在消费者中的实用性。