A Short Study to Assess the Potential of Independent Component Analysis for Motion

评估独立成分分析在运动干扰去除中的潜能的一个小研究

Proceedings of the 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference Shanghai, China, September 1-4, 2005 A Short Study to Assess the Potential of Independent Component Analysis for Motion Artifact Separation in Wearable Pulse Oximeter Signals Jianchu Yao1 and Steve Warren2 Department of Technology Systems, East Carolina University, Greenville, NC, USA 2 Department of Electrical & Computer Engineering, Kansas State University, Manhattan, KS, USA 1 Abstract—Motion artifact reduction and separation become critical when medical sensors are used in wearable monitoring scenarios. Previous research has demonstrated that independent component analysis (ICA) can be applied to pulse oximeter signals to separate photoplethysmographic (PPG) data from motion artifacts, ambient light, and other interference in lowmotion environments. However, ICA assumes that all source signal component pairs are mutually independent. It is important to assess the statistical independence of the source components in PPG data, especially if ICA is to be applied in ambulatory monitoring environments, where motion artifacts can have a substantial effect on the quality of data received from lightbased sensors. This paper addresses the statistical relationship between motion artifacts and PPG data by calculating the correlation coefficients between arterial volume variations and motion over a range of stationary to high-motion conditions. Analyses indicate that motion significantly affects arterial flow, so care must be taken when applying ICA to light-based sensor data acquired from wearable platforms. Keywords—Motion artifacts, independent analysis, pulse oximetry, wearable sensors component pulse oximeters: arterial volume variation and motion artifact. The following two sections discuss the promise of ICA in pulse oximetry and provide the theoretical basis for this approach. The Methods section then describes how data were collected and how the statistical independence between arterial volume variation and motion was assessed. The Results section summarizes these analyses and notes that the application of ICA to PPG data should be performed with caution. II. BACKGROUND AND MOTIVATION Independent component analysis is usually denoted as x(t ) = M ⋅ s (t ) where (1) M ∈R m×n is the mixing matrix, x(t ) ∈ R n m is the I. INTRODUCTION Wearable medical devices are becoming popular in home healthcare, telemedicine, rehabilitation, and athletic training due to advances in sensors, device miniaturization, power consumption, computation speed, and wireless communication. These devices surpass the usefulness of their traditional desktop counterparts by continuously monitoring vital signs in “real-world” usage scenarios. However, motion artifact reduction and separation are a more serious concern in wearable environments, prompting the attention of researchers [1-5]. Recently, increased emphasis has been placed on Independent Component Analysis (ICA) for the separation of motion artifacts from desired quantities [1, 2, 6, 7]. ICA is especially attractive because it does not require prior knowledge of the system. (ICA is consequently referred to as blind source separation.) ICA methods can separate mixed signals with multiple source components when multiple observation sets can be acquired. Application of ICA is based on the assumption that all source signal components are mutually independent. It is therefore necessary to examine the relationship between each source signal pair to make sure that they satisfy this prerequisite. This paper addresses the statistical independence of two signal sources in photoplethysmographic (PPG) observed mixed signal vector, and s (t ) ∈ R is the source component vector. This approach has been applied to the well-known “cocktail party” problem, where multiple (n) speakers’ (singers’) voices are recorded (mixed) by multiple (m) microphones. The task of ICA is to recover each speaker’s original voice s (t ) from the mixture x (t ) . ICA has found application in the biomedical field, having been successfully utilized in human electroencephalograms [8, 9] and functional magnetic resonance imaging [10]. Recently, researchers realized that ICA might be suitable for the reduction and separation of both motion artifacts and ambient light interference in photoplethysmographic pulse oximeter data [1, 2, 6, 7]. In this context, s (t ) ∈ R refers to the n source signal components: the volume fraction of arterial blood, venous blood, bones, and other non-blood tissues. The mixing matrix, M , represents the attenuation of multi-wavelength optical signals by the n fractional volumes when traveling through blood and tissues. Finally, the mixture, x (t ) , is the m signals received from the optical detector corresponding to the m wavelengths. In a real system, the mixture is usually a current signal that has been conditioned with analog and digital processing elements. n In practice, the number of independent components contributing to the mixture signal is unclear. A typical optical pulse oximeter can, with minimal hardware and software This material is based upon work supported by the National Science Foundation under grants BES–0093916. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the NSF. 0-7803-8740-6/05/$20.00 ©2005 IEEE. 3585

第1页下一页