房颤智能检测中的心电特征选择和机器学习

房颤是最常见的心律失常疾病,因其临床诊断率低而促进了实时自动检测算法的发展.但大多算法缺乏模型或数据库间的对比,难以评判模型的优劣.为此考虑选择3种机器学习算法(支持向量机、随机森林、逻辑回归)构建3个独立的房颤检测模型.3种模型分别在MIT-BIH房颤数据库上训练,并在3个独立数据库上进行测试和对比,同时进一步分析特...     

QRS Detection Based on Multiscale Mathematical Morphology for Wearable ECG Devices in Body Area Networks

A novel wearable electrocardiograph (ECG) QRS detection algorithm for wearable ECG devices in body area networks is presented in this paper, which utilizes the multistage multiscale mathematical morphology filtering to suppress the impulsive noise and uses the multiframe differential modulus accumulation to remove the baseline drift and enhance the signal. The proposed algorithm, verified with data from the MIT/BIH Arrhythmia Database and wearable ECG devices, achieves an average QRS detection rate of 99.61%, a sensitivity of 99.81%, and a positive prediction of 99.80%. It compares favorably to the published methods.      

ECG去噪算法的设计与实现

对心电信号进行数字滤波时,因其幅频特性和过渡带的影响很容易造成心电波形失真,为了使原始含噪心电信号与降噪后的心电信号具有同等光滑性和最小方差估计特性,去噪连续性效果更好,分析研究了几种常用的去噪算法,并分别采用MIT BIH数据库中的数据及实测心电数据进行仿真分析,仿真结果表明采用小波阈值去噪算法可以更好的对心电信号进行去噪处理,得到清晰不失真的原始心电信号,信噪比(SNR)达到30以上,均方误差（MSE）在0.1以下,设计方法简单、计算量小,具有很好的应用前景和广泛地使用价值。     

基于区域极值法心电QRS波的检测

目前,心电QRS波检测的方法很多,其中小波变换较为流行。其核心策略通常是在某一尺度或某几个尺度内搜索小波变换模极大—极小值对之间的过零点,此点对应R波峰值位置。然而,此方法策略存在一定的局限性,它只对一部分病例具有较好的检测效果,对于其它病例则不能较好地检测到R波峰值点,存在不同程度的偏差。根据医学心电常识,R波在心电信号一定区域内幅值为最大,只要确定了搜索区域,搜索到极大值点即为R波峰值点。本文结合二次B-样条小波变换算法提出一种区域极值检测法。通过美国MIT/BIH心电标准数据库进行实验验证表明,该方法克服了此局限性,对108和207以外的所有病例都具有较高的检测率和较好的检测效果,同时经过ARM平台实际应用验证了本算法完全满足实时分析的需求。     

一种心电信号压缩算法——代表波形法

本文根据国际ＳＣＰ－ＥＣＧ规定的标准,对心电信号的压缩技术进行了研究。首先研究了ＱＲＳ波的识别算法,对心电信号的滤波技术进行了深入的探讨,设计了多个性能优越,高速节时的滤波器,采用自适应阈值检测技术和双阈值检测设计出一种高效、可靠的ＱＲＳ波检测算法;运用信号平均技术,产生具有高精度的代表波形;对残差信号处理后进行压缩编号,最后得到一种比较理想的心电信号压缩算法。     

Selective integration of local‐feature detector by boosting for pedestrian detection

An example‐based classification algorithm for pedestrian detection is presented. The classifier integrates component‐based classifiers according to the AdaBoost algorithm. A probability estimate by a kernel‐SVM is used for the outputs of base learners, which are independently trained for local features. The base learners are determined by selecting the optimal local feature according to sample weights determined by the boosting algorithm with cross‐validation. Our method was applied to the MIT CBCL pedestrian image database, and 54 subregions were extracted from each image as local features. The experimental results showed a good classification ratio for unlearned samples. © 2011 Wiley Periodicals, Inc. Electr Eng Jpn, 177(4): 12–22, 2011; Published online in Wiley Online Library ( wileyonlinelibrary.com ). DOI 10.1002/eej.21195     

Robustness study of ECG biometric identification in heart rate variability conditions

Electrocardiogram (ECG) has been actively proposed as an aliveness biometric for years. In realistic application, this type of biometrics still needs to be verified in some conditions related to the practical use. Therefore, we propose in this paper the study of two themes that concern its realistic application. First, the single beat of an electrocardiogram (SB‐ECG) is used as the ECG feature, which aims to decrease the time to acquire the ECG from the individual. Second, for the robustness study of ECG features to heart rate variability (HRV), we investigate the relation between the correct identification and the HRV conditions. The HRV‐ECG data is set up and collected by using the Bio‐Pac system. The overall processes of the proposed method are the following: Continuous wavelet transform (CWT) is used to analyze each SB‐ECG. Then the root‐mean square (RMS) value of total energy of the wavelet coefficients of each P, QRS, and T segment is calculated. Next, the Fisher linear discriminant analysis (FLDA) is applied to all sets of the RMS value for dimension reduction. Lastly, the normalized Euclidean distance is implemented as the classifier. The experimental results demonstrate that, with the proposed method, the classification achieves 97% accuracy and the robustness study achieves over 80% accuracy with HRV‐ECG change up to 20%. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

A novel DWT method for ECG noise elimination

An electrocardiogram (ECG) often contains various types of noises and artifacts that might lead to wrong analysis. Recently, many techniques based on discrete wavelet transform (DWT) for ECG noise elimination have been proposed. Determination of the number of decomposition levels, which could vary with the sampling rate (frequency sampling), is one of the main issues in DWT. This letter presents an automatic index, called mean power frequency (MPF) and is independent of the sampling rate, for stopping decomposing process when it achieves the optimum number of decomposition levels. The effectiveness of this scheme is expressed by the signal‐to‐noise ratio (SNR), mean square error (MSE), and correlation coefficient (CC) between the pure and corrected ECG. The results indicate that the applied method can remove Gaussian noise efficiently. © 2015 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

无线脑电睡眠分期系统设计及算法研究

对睡眠进行分期的关键要求是不失真的地采集到脑电信号和快速有效的分期算法。针对这一问题,设计出了便携式无线脑电信号采集器,并提出了使用样本熵-小波分析(Samp En-Wavelet)方法进行睡眠各阶段分期。脑电信号采用Wi Fi方式传输数据,采样率250 Hz~2 k Hz可调,通过对MIT-BIH睡眠脑电数据库中使用人工分期方法的对比,分期结果相同率达到了80%以上,简化了RK和AASM人工睡眠多参数分期方法的繁琐工作。5名被试的分期结果表明,本系统与专业睡眠监测设备分期结果相近(82%的重复度)。这一低成本、便携式的睡眠分期系统在睡眠分期研究及普及应用方面具有重要的价值?     

基于 MIC 心率变异性特征选择的情感识别研究

心率变异性分析能够在情感识别中发挥重要作用,为了建立心电与情感类别之间的精准模型,提出了基于最大信息系 数(maximal information coefficient,MIC)的特征选择方法。 使用 Aubt 数据库和设计情感诱发实验进行研究,首先提取了心率变 异性时域、频域、非线性及时频域 40 个特征参数,然后基于 MIC 方法结合支持向量机、随机森林、K 近邻算法进行情感建模。 结 果显示,基于 MIC 特征选择方法,使用 Aubt 数据库针对唤醒度、效价、4 类情感的分类准确度分别为 90%、89%、84%,并进一步 选用皮尔森相关系数、ANOVA 特征选择方法与 MIC 进行对比;诱发实验数据中的多种一对一情感识别率均高于 75%。 结果表 明基于 MIC 特征选择方法能够显著提高分类准确度,对基于心电信号进行情感识别具有重要意义。     

Search of target image area by using geometrical and gray‐level features

This paper proposes a method to measure the two‐dimensional pose of an object by using voting process based on geometrical and gray‐level features between edge pixels and the base point. The proposed method can measure a location and an inclination of an object even if the object has a free pose in an image. In the template registration process, the geometrical features and the distribution of gray levels for all edge pixels are registered into the template table. In the matching process, not only the geometrical features but also the arrays of gray levels between the edge pixels and the base point are used for the matching. The experimental results show that the poses of objects with free locations and inclinations were recognized successfully by the proposed method. High‐speed matching was realized in comparison with the conventional methods. © 2007 Wiley Periodicals, Inc. Electr Eng Jpn, 161(4): 12– 19, 2007; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/eej.20613     

Improvement of automatic Chinese text classification by combining multiple features

In this paper, we present an effective way of combining character‐based (N‐gram) and word‐based approaches for Chinese text classification. Uni‐gram and bi‐gram features are considered as the baseline model, which are then combined with word features of length greater than or equal to 3. A weight coefficient that can be used to give higher weights to word features is also introduced. We further employ a serial approach based on feature transformation and dimension reduction techniques. The results of McNemar's test indicate that the performance is significantly improved by our proposed method. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

Recognizing sentiment of relations between entities in text

Recently, sentiment analysis for identifying positive or negative opinions from texts has received much attention. In this paper, we introduce sentiment analysis into a new field, which recognizes sentiment of relations between entities in the text. Three sentiment polarities between entities are recognized, namely positive, negative, and neutral. The difficulty in this work is that several pairs of entities may appear in the same sentence, and their sentiment polarities are determined by different related regions of the sentence. In addition, different features of words and their interactions in a related region will affect the final sentiment. It is difficult to process this using rigid rule‐based methods. Therefore, we propose a machine‐learning method based on statistics. In the proposed method, the model of conditional random fields (CRFs) is used to annotate the sentiment polarity between entities with the help of the syntactic dependency tree. The string of words that connects two entities in the dependency tree is used as the related region to recognize the sentiment. Experimental results and comparison with the other methods based on different principles and related regions suggest that the proposed method shows better performance and proves its validity. Moreover, the effect of different features on the CRF is word of the ith word in the sentence. We can get a CRF model using the CRF++ tool based on feature template and training corpus. After obtaining the model, sentiment of relations could be assigned automatically. The algorithm is shown in Fig. 2. © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

Association Rule Mining for Continuous Attributes using Genetic Network Programming

Most of the existing association rule mining algorithms are able to extract knowledge from databases with attributes of binary values. However, in real‐world applications, databases are usually composed of continuous values such as height, length or weight. If the attributes are continuous, the algorithms are commonly integrated with a discretization method that transforms them into discrete attributes. Discretization is a process of transforming a continuous attribute value into a finite number of intervals and assigning each interval into a discrete numerical value. However, the user most often must specify the number of intervals, or provide some heuristic rules to be used while discretization, and then it is difficult to get the highest attribute interdependency and at the same time get the lowest number of intervals. In this paper we present an association rule mining algorithm that is suited for continuous valued attributes commonly found in scientific and statistical databases. We propose a method using a new graph‐based evolutionary algorithm named ‘genetic network programming (GNP)’ that can deal with continuous values directly, that is, without using any discretization method as a preprocessing step. GNP represents its individuals using graph structures and evolves them in order to find a solution; this feature contributes to creating very compact programs and implicitly memorizing past action sequences. In the proposed method using GNP, the significance of the extracted association rules is measured by the use of χ² test, and only important association rules are stored in a pool all together through generations. Results of experiments conducted on a real‐life database suggest that the proposed method provides an effective technique for handling continuous attributes. Copyright © 2008 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.     

Poincare mapping for detecting abnormal dynamics of cardiacrepolarization

This article concentrates on analysis of subtle ECG signal features associated with characteristic temporal variations in the repolarization phase of cardiac electrical activation (i.e., variations that are embedded within the T-wave). These variations are termed T-wave alternans (TWA) to connote larger similarity between T-wave shapes in every other beat than that in the adjacent beats that occur during regular heart rhythm of an increased rate. In an early work the concept of associating TWA with period-doubling bifurcation of the cardiac oscillator has been put forward and demonstrated on canine ECG traces. In this article we extend this idea and use it for quantification of TWA in human ECGs recorded from postinfarction patients. The purpose of this work is to show that: a statistically significant correlation exists for the analyzed ECGs between the TWA level computed by means of an inter-cycle synchronized sampling technique known as Poincare mapping (PM) and a widely used Fourier spectrum (FS) method, and that the PM method outperforms the widely used FS method for TWA analysis in a number of ways     

Analysis and Implementation of an Interleaved ZVS Converter with High Input Voltage

An interleaved half‐bridge converter is presented for high input voltage application. The features of the proposed converter are zero voltage switching (ZVS) turn‐on for all active switches, ripple current reduction at output side, load current sharing and load voltage regulation. Two half‐bridge converters connected in series and two split capacitors are used to limit the voltage stress of each power switch at one‐half of input DC bus voltage. Thus, active switches with low voltage stress can be used at high input voltage application. On the other hand, the output sides of two half‐bridge converters are connected in parallel to share the load current and reduce the current stresses of the secondary windings and the rectifier diodes. Since two half‐bridge converters are operated with interleaved pulse‐width modulation (PWM), the output ripple current can partially cancel each other such that the resultant ripple current at output side is reduced and the size of output inductors can be reduced. In each half‐bridge converter, asymmetrical PWM scheme is used to regulate the output voltage. Based on the resonant behavior by the output capacitance of MOSFETs and the leakage inductance (or external inductance) of transformers, active switches can be turned on at ZVS during the transition interval. Thus, the switching losses of power MOSFETs are reduced. The proposed converter can be applied for high input voltage applications such as three‐phase 380‐V utility system. Finally, experiments based on a laboratory prototype with 960‐W rated power are provided to demonstrate the performance of proposed converter. Copyright © 2012 John Wiley & Sons, Ltd.     

Design of efficient fractional operator for ECG signal detection in implantable cardiac pacemaker systems

A low power and high-performance digital electrocardiogram (ECG) detector has become a basic requirement in modern implantable cardiac pacemakers. A fractional operator-based digital ECG detector for modern pacemaker systems is proposed in this work. Instead of conventional thresholding, an adaptive slope prediction threshold is utilized for the detection of ECG peaks. A stochastic search-based algorithm, namely, cuckoo search algorithm, is used to design an optimal fractional operator that is used for ECG denoising. It has been found that the proposed adaptive slope prediction threshold increases the QRS complex detection performance. A low detection error rate (DER) ranges from 0.01% to 0.56%, positive predictivity (P+) ranges from 99.32% to 99.98%, sensitivity (Se) ranges from 99.45% to 99.98%, and a detection accuracy (Acc) ranges from 99.43% to 99.96% for different databases are achieved for the proposed ECG detector, which is better compared with the existing ECG detectors. The proposed design of fractional order operator based on the lattice wave digital filter (LWDF) requires a minimum number of the multipliers for its structural realization.     

An accurate forearm EMG signal classification method using two‐channel electrode

An accurate electromyography (EMG) classification algorithm to control a virtual hand prosthesis with 12 degrees of freedom using two surface EMG electrodes is presented in this paper. We propose the application of independent component analysis (ICA) for blind‐source separation of the EMG signals obtained from two electrodes. One of the problems affecting the EMG classification accuracy is the location dependence of the EMG signal due to the superposition of signals from multiple sources. ICA is used to separate the two signals obtained from two surface electrodes into two independent EMG signals prior to the feature extraction and classification processes. We demonstrate that the EMG classification accuracy can be improved using the ICA algorithm. We also propose a novel eigen‐based feature that is extracted from the short‐time Fourier transform (STFT) magnitude spectrum. Our new feature not only decreases feature dimensions but also performs better than other well‐known features. We also implement the EMG classification scheme on the virtual robot arm. The performance shows promising result as indicated by a decrease in the Davies–Bolden (DB) index after applying the ICA © 2013 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.