A novel approach to predict energy of physical activity

描述（由申请人提供）：体力活动（PA）在许多慢性疾病中的作用越来越受到认可。PA的准确和详细的测量是进一步探索其与健康和疾病的关系的重要前提。小型和可穿戴的加速度计允许PA的客观测量（PA计数），同时还提供与PA相关的能量消耗的粗略估计（EEAcT）。然而，目前的PA监测器仅限于使用时间平均信号和线性回归算法，这些算法始终提供不准确的EEAc-r预测-PA强度的关键特征。为了从根本上指导PA监视器的未来设计以准确预测EEAcT，我们假设可以从多个身体部分的原始加速度和姿势信号中提取参数。使用先进的仪器和技术专业知识的独特组合，我们建议开发一种新的分析方法，用于准确预测EEAc重新利用来自上半身和下半身的原始加速度信号。这是通过使用由10个加速度计阵列组成的定制设计的监视器以32个样本/秒的速率连续测量运动和姿势来实现的。我们将使用全房间间接热量计测量24小时内的每分钟EEAcT，并使用便携式热量计测量3小时的自由生活时间。该测量的EEAcT将用作预测模型的目标。我们将采用先进的建模技术（人工神经网络）来模拟提取的PA参数，以达到准确的预测EEAcT。重复测量将用于交叉验证模型的预测准确性。该研究旨在涵盖肥胖、超重和消瘦成人的异质性人群样本（n = 200），并包括广泛的PA类型和强度。这项研究的意义在于，我们的研究结果将为开发下一代PA监测器提供见解，例如传感器应该放置在身体上的哪个位置，应该提取什么信号参数，以及应该如何应用分析算法。此外，我们的研究将改善和验证EEAcT预测几个市场上可用的PA监测器，从而提供直接的好处，他们在该领域的应用。

DESCRIPTION (provided by applicant): The role of physical activity (PA) in many chronic diseases is increasingly being recognized. The accurate and detailed measurement of PA is a crucial prerequisite to further explore its association with health and disease. Small and wearable accelerometers allow objective measurement of PA (PA counts), while also providing a rough estimation of the energy expenditure associated with PA (EEAcT). However, current PA monitors are restricted to using time-averaged signals and linear regression algorithms which consistently provide inaccurate predictions of EEAc-r- the key characteristic of PA intensity. To fundamentally guide the future designs of PA monitors to accurately predict EEAcT, we hypothesize that parameters can be extracted from the raw acceleration and postural signals of multiple body segments. Using a unique combination of sophisticated instruments and technical expertise, we propose to develop a novel analytical approach for accurately predicting EEAc-reutilizing the raw acceleration signals from upper and lower body segments. This is accomplished by continuously measuring movement and postures, at a rate of 32 samples/second, using a custom-designed monitor that consists of an array of 10 accelerometers. We will measure minute-to minute EEAcT using a whole-room indirect calorimeter for a 24-hour period, and a portable calorimeter for a 3-hour free-living period. This measured EEAcT will be used as the target for the prediction model. We will apply an advanced modeling technique (artificial neural networks) to model the extracted PA parameters to arrive at an accurate prediction of EEAcT. Repeated measurements will be used to cross-validate the prediction accuracy of the model. The study is designed to encompass a heterogeneous population sample (n=200) of obese, overweight, and lean adults, and to include a wide range of PA types and intensities. The significance of this research is that our results will provide insight for developing the next-generation PA monitors, such as where on the body the sensors should be placed, what signal parameters should be extracted, and how the analytical algorithms should be applied. In addition, our study will improve and validate EEAcT prediction by several market-available PA monitors, thus offering immediate benefits to their applications in the field.