Robust supervised classification of motor unit action potentials

A certainty-based classification algorithm is described, which comprises part of a clinically used EMG signal decomposition system. This algorithm classifies a candidate motor unit action potential (MUAP) to the motor unit action potential train (MUAPT) that produces the greatest estimated certainty, provided this maximal certainty is above a given threshold. The algorithm is iterative, such that the certainty with which assignments are made increases with each pass through the data, and it has specific stopping criteria. The performance and sensitivity (to the assignment threshold) of the Certainty algorithm and an iterative minimum Euclidean distance (MED) algorithm are compared by classifying sets of MUAPs detected in real concentric needle-detected EMG signals, using a range of assignment thresholds for each algorithm. With regard to MUAP assignment and error rates, the Certainty algorithm consistently provides better mean results and, more importantly, less variable results than the MED algorithm. The Certainty algorithm can provide mean assignment and error rates of 80.8 and 1.5%, respectively, with a maximum error rate of 3.2%; the MED algorithm can provide mean assignment and error rates of 80.3 and 3.3%, respectively, with a maximum error rate of 6.5%. The Certainty algorithm is relatively insensitive to the certainty threshold used, can consistently differentiate between similarly shaped MUAPs from different MUAPTs, and can make correct classifications despite biological shape variability, background noise and signal shape non-stationarity.     

Adaptive motor unit action potential clustering using shape and temporal information

An adaptive algorithm is described that groups motor unit action potentials (MUAPs), detected in a composite EMG signal during signal decomposition, and creates partial motor unit action potential trains (MUAPTs). Data-driven MUAP shape and motor unit firing-pattern based criteria are used to form the clusters. An algorithm for estimating MUAPT temporal parameter, which provides accurate estimates even for partially defined trains, is used to obtain firing-pattern information. No a priori knowledge is required regarding the number of clusters or the distribution of their template shapes. The clustering algorithm when applied to real concentric-needle detected MUAP data provides accurate and useful clustering results. Compared to a classical leader-based algorithm, it provides more robust performance, is better able to estimate the true number of motor units represented in a set of detected MUAPs, and obtains more complete and accurate MUAPTs.     

A threshold fuzzy entropy based feature selection for medical database classification

Feature selection is one of the most common and critical tasks in database classification. It reduces the computational cost by removing insignificant features. Consequently, this makes the diagnosis process accurate and comprehensible. This paper presents the measurement of feature relevance based on fuzzy entropy, tested with a Radial Basis Function Network classifier for a medical database classification. Three feature selection strategies are devised to obtain the valuable subset of relevant features. Five benchmarked datasets, which are available in the UCI Machine Learning Repository, have been used in this work. The classification accuracy shows that the proposed method is capable of producing good results with fewer features than the original datasets.     

Feature dimensionality reduction for myoelectric pattern recognition: A comparison study of feature selection and feature projection methods

This study investigates the effect of the feature dimensionality reduction strategies on the classification of surface electromyography (EMG) signals toward developing a practical myoelectric control system. Two dimensionality reduction strategies, feature selection and feature projection, were tested on both EMG feature sets, respectively. A feature selection based myoelectric pattern recognition system was introduced to select the features by eliminating the redundant features of EMG recordings instead of directly choosing a subset of EMG channels. The Markov random field (MRF) method and a forward orthogonal search algorithm were employed to evaluate the contribution of each individual feature to the classification, respectively. Our results from 15 healthy subjects indicate that, with a feature selection analysis, independent of the type of feature set, across all subjects high overall accuracies can be achieved in classification of seven different forearm motions with a small number of top ranked original EMG features obtained from the forearm muscles (average overall classification accuracy >95% with 12 selected EMG features). Compared to various feature dimensionality reduction techniques in myoelectric pattern recognition, the proposed filter-based feature selection approach is independent of the type of classification algorithms and features, which can effectively reduce the redundant information not only across different channels, but also cross different features in the same channel. This may enable robust EMG feature dimensionality reduction without needing to change ongoing, practical use of classification algorithms, an important step toward clinical utility.     

A morphological analysis of the macro motor unit potential

The technique of macro EMG is used to investigate the motor unit architecture in a number of pathological conditions. Amplitude and area are the most commonly used criteria, but these parameters alone are not sufficient to assess the complexity of the macro MUP morphology. In an attempt to examine the morphology of the macro MUP in more detail, additional measures were investigated including, (i) average power, (ii) duration, and (iii) number of phases. Macro MUP duration was defined as the time parameter that contains a particular fraction (90%) of the total power of the potential. The above mentioned parameters were evaluated for normal subjects and for patients suffering with motor neuron disease (MND), spinal muscular atrophy (SMA), and Becker's muscular dystrophy (BMD). It is shown that high amplitude and average power macro MUPs give shorter macro MUP duration than macro MUPs with normal amplitude. In contrast, in low amplitude macro MUPs there is a tendency towards a higher duration measure, as compared with the duration of the normal amplitude macro MUPs. Also, t–test results for the duration measure gave a significant difference between the NOR–MND, and no significant difference between the NOR–BMD and NOR–SMA groups at P<0.05. Significant difference between the NOR and the three disease groups investigated was obtained for the parameters log amplitude, log area, and log average power. The number of phases was not significantly different between the NOR and the rest of the groups. In conclusion, the average power and duration parameters can possibly be used as additional discriminators to detect abnormalities of the macro motor unit potential in both needle and surface EMG but further investigation is necessary.     

Robust method for estimating motor unit firing-pattern statistics

An error-filtered estimation (EFE) algorithm for estimating the mean and standard deviation of a set of time intervals between consecutive motor unit firing times (inter-pulse intervals (IPIs)) is described. As the input IPI data are filtered and only valid IPIs are used to estimate mean and standard deviation values, the EFE algorithm provides accurate estimates even when the data defining the train of motor unit firing times are only partially complete or have several erroneous firing times. The algorithm has been evaluated using both simulated and real motor unit firing time data, and has been found to provide accurate and unbiased mean and standard deviation estimates, even when up to 70% of the IPI data are incorrect.     

Resolving superimposed motor unit action potentials

A new algorithm to resolve superimposed motor unit action potentials (MUAPs) is described, which uses a reduced search space and is based on the peel off approach. Knowledge specific to the problem domain, such as temporal relationships between and within motor unit action potential trains and MUAP energy information, is used to reduce the search space of motor units, possibly contributing to a superposition. The algorithm is tested using real electromyographic signals, and it demonstrates robust performance across the signals tested. For the signals tested, the average total resolution rate is 94%, the average correct resolution rate is 99.2% and the average error rate is 0.85%.     

Optimal feature selection for the assessment of vocal fold disorders

Unilateral vocal fold paralysis, vocal fold polyp, and vocal fold nodules are the most common types of neurogenic and organic vocal disorders. This article aims to distinguish these types of vocal diseases into four different classes for the purpose of automatic screening. Firstly, the reconstructed signal at each wavelet packet decomposition sub-band in five levels of decomposition with mother wavelet of (db10) is used to extract the nonlinear features of self-similarity and approximate entropy. Also, wavelet packet coefficients are used to measure energy and Shannon entropy features at different spectral sub-bands. Consequently, to find a discriminant feature vector, three different methods have been applied: Davies-Bouldin (DB) criteria, genetic algorithm (GA) with the fitness functions of support vector machine's (SVM) and k-nearest neighbor's (KNN) recognition rates. Finally, obtained feature vectors have been passed on to SVM and KNN classifiers. The results show that a feature vector of length 12 obtained by the optimization method of GA with the fitness function of SVM's recognition rate fed to SVM classifier achieves the highest classification accuracy of 91%. Furthermore, nonlinear features play an important role in pathological voice classification by participating rate of approximately 67% in the optimal feature vector.     

Motor unit action potential rate and motor unit action potential shape properties in subjects with work-related chronic pain

The objective of this study was to investigate differences in motor control of the trapezius muscle in cases with work-related chronic pain, compared to healthy controls. Ten cases with chronic pain and 13 controls participated in the study. Electromyographic (EMG) signals were recorded from the upper trapezius during five computer work-related tasks. Motor control was assessed using global root-mean-square value (RMS_G), motor unit action potential (MUAP) rate (number of MUAPs per second, MR) and two MUAP shape parameters, i.e. root-mean-square (RMS_MUAP) and median frequency (FMED_MUAP). MR and FMED_MUAP were higher for the cases than for the controls (P<0.05). RMS_MUAP showed a trend for higher values in the chronic pain group (P<0.13), whereas RMS_G did not show a significant difference between the groups. The higher MR, FMED_MUAP and the trend for higher RMS_MUAP suggest that more high-threshold MUs contribute to low-level computer work-related tasks in chronic pain cases. Additionally, the results suggest that the input of the central nervous system to the muscle is higher in the cases with chronic pain.     

Intensity based affine registration including feature similarity for spatial normalization

This paper presents a new spatial normalization with affine transformation. The quantitative comparison of brain architecture across different subjects requires a common coordinate system. For the analysis of a specific brain area, it is required to normalize and compare a region of interest and global brain. Intensity based registration method matches the global brain well. But a region of interest may not be locally normalized compared to feature based method. The method of this paper uses feature similarities of local region as well as intensity similarities. The lateral ventricle and the central gray nuclei of brain including the corpus callosum, which is used for features in Schizophrenia detection, is appropriately normalized. In the results section, our method reduces the difference of feature area such as corpus callosum (7.7%, 2.4%) and lateral ventricle (8.2%, 13.5%) compared with mutual information and Talairach methods.     

GMDH-based feature ranking and selection for improved classification of medical data

Medical applications are often characterized by a large number of disease markers and a relatively small number of data records. We demonstrate that complete feature ranking followed by selection can lead to appreciable reductions in data dimensionality, with significant improvements in the implementation and performance of classifiers for medical diagnosis. We describe a novel approach for ranking all features according to their predictive quality using properties unique to learning algorithms based on the group method of data handling (GMDH). An abductive network training algorithm is repeatedly used to select groups of optimum predictors from the feature set at gradually increasing levels of model complexity specified by the user. Groups selected earlier are better predictors. The process is then repeated to rank features within individual groups. The resulting full feature ranking can be used to determine the optimum feature subset by starting at the top of the list and progressively including more features until the classification error rate on an out-of-sample evaluation set starts to increase due to overfitting. The approach is demonstrated on two medical diagnosis datasets (breast cancer and heart disease) and comparisons are made with other feature ranking and selection methods. Receiver operating characteristics (ROC) analysis is used to compare classifier performance. At default model complexity, dimensionality reduction of 22 and 54% could be achieved for the breast cancer and heart disease data, respectively, leading to improvements in the overall classification performance. For both datasets, considerable dimensionality reduction introduced no significant reduction in the area under the ROC curve. GMDH-based feature selection results have also proved effective with neural network classifiers.     

一种面向生物信息学数据的模式选择算法研究

本文面向生物信息学中一类重要问题———模式选择问题展开研究。针对模式选择过程中,算法复杂度高以及最佳模式量个数难以确定的问题,提出一种基于互信息(MI)理论实现模式选择,基于模糊神经的模式子集评价准则实现最佳模式量选择的算法。该算法基于模式信息与类别信息之间的相关程度,以及各子模式之间的冗余程度实现模式选择;基于模糊模式指标完成特征模式子集评价。实验数据采用数据挖掘后的小鼠基因表达数据(来自Leiden University)与UCI数据。结果表明,算法性能优良,无论在复杂度还是正确率方面效果均有所提高。     

EMG feature assessment for myoelectric pattern recognition and channel selection: A study with incomplete spinal cord injury

Myoelectric pattern recognition with a large number of electromyogram (EMG) channels provides an approach to assessing motor control information available from the recorded muscles. In order to develop a practical myoelectric control system, a feature dependent channel reduction method was developed in this study to determine a small number of EMG channels for myoelectric pattern recognition analysis. The method selects appropriate raw EMG features for classification of different movements, using the minimum Redundancy Maximum Relevance (mRMR) and the Markov random field (MRF) methods to rank a large number of EMG features, respectively. A k-nearest neighbor (KNN) classifier was used to evaluate the performance of the selected features in terms of classification accuracy. The method was tested using 57 channels’ surface EMG signals recorded from forearm and hand muscles of individuals with incomplete spinal cord injury (SCI). Our results demonstrate that appropriate selection of a small number of raw EMG features from different recording channels resulted in similar high classification accuracies as achieved by using all the EMG channels or features. Compared with the conventional sequential forward selection (SFS) method, the feature dependent method does not require repeated classifier implementation. It can effectively reduce redundant information not only cross different channels, but also cross different features in the same channel. Such hybrid feature-channel selection from a large number of EMG recording channels can reduce computational cost for implementation of a myoelectric pattern recognition based control system.     

Evaluation of Karhunen-Loeve expansion for feature selection in computer-assisted classification of bioprosthetic heart-valve status

This paper analyses the performance of four different feature-selection approaches of the Karhunen-Loève expansion (KLE) method to select the most discriminant set of features for computer-assisted classification of bioprosthetic heart-valve status. First, an evaluation test reducing the number of initial features while maintaining the performance of the original classifier is developed. Secondly, the effectiveness of the classification in a simulated practical situation where a new sample has to be classified is estimated with a validation test. Results from both tests applied to a reference database show that the most efficient feature selection and classification (> or = 97% of correct classifications (CCs)) are performed by the Kittler and Young approach. For the clinical databases, this approach provides poor classification results for simulated 'new samples' (between 50 and 69% of CCs). For both the evaluation and the validation tests, only the Heydorn and Tou approach provides classification results comparable with those of the original classifier (a difference always < or = 7%). However, the degree of feature reduction is particularly variable. The study demonstrates that the KLE feature-selection approaches are highly population-dependent. It also shows that the validation method proposed is advantageous in clinical applications where the data collection is difficult to perform.     

A simulation to analyze feature selection methods utilizing gene ontology for gene expression classification

Gene expression profile classification is a pivotal research domain assisting in the transformation from traditional to personalized medicine. A major challenge associated with gene expression data classification is the small number of samples relative to the large number of genes. To address this problem, researchers have devised various feature selection algorithms to reduce the number of genes. Recent studies have been experimenting with the use of semantic similarity between genes in Gene Ontology (GO) as a method to improve feature selection. While there are few studies that discuss how to use GO for feature selection, there is no simulation study that addresses when to use GO-based feature selection. To investigate this, we developed a novel simulation, which generates binary class datasets, where the differentially expressed genes between two classes have some underlying relationship in GO. This allows us to investigate the effects of various factors such as the relative connectedness of the underlying genes in GO, the mean magnitude of separation between differentially expressed genes denoted by δ, and the number of training samples. Our simulation results suggest that the connectedness in GO of the differentially expressed genes for a biological condition is the primary factor for determining the efficacy of GO-based feature selection. In particular, as the connectedness of differentially expressed genes increases, the classification accuracy improvement increases. To quantify this notion of connectedness, we defined a measure called Biological Condition Annotation Level BCAL(G), where G is a graph of differentially expressed genes. Our main conclusions with respect to GO-based feature selection are the following: (1) it increases classification accuracy when BCAL(G) ⩾ 0.696; (2) it decreases classification accuracy when BCAL(G) ⩽ 0.389; (3) it provides marginal accuracy improvement when 0.389 < BCAL(G) < 0.696 and δ < 1; (4) as the number of genes in a biological condition increases beyond 50 and δ ⩾ 0.7, the improvement from GO-based feature selection decreases; and (5) we recommend not using GO-based feature selection when a biological condition has less than ten genes. Our results are derived from datasets preprocessed using RMA (Robust Multi-array Average), cases where δ is between 0.3 and 2.5, and training sample sizes between 20 and 200, therefore our conclusions are limited to these specifications. Overall, this simulation is innovative and addresses the question of when SoFoCles-style feature selection should be used for classification instead of statistical-based ranking measures.     

Adaptive certainty-based classification for decomposition of EMG signals

An adaptive certainty-based supervised classification approach for electromyographic (EMG) signal decomposition is presented and evaluated. Similarity criterion used for grouping motor unit potentials (MUPs) is based on a combination of MUP shapes and two modes of use of motor unit (MU) firing pattern information: passive and active. Performance of the developed classifier was evaluated using synthetic signals of known properties and real signals and compared with the performance of the certainty classifier (CC). Across the sets of simulated and real EMG signals used for comparison, the adaptive certainty classifier (ACC) had both better average performance and lower performance variability. For simulated signals of varying intensity, the ACC had an average correct classification rate (CC _r ) of 83.7% with a mean absolute deviation (MAD) of 5.8% compared to 78.3 and 8.7%, respectively, for the CC. For simulated signals with varying amounts of shape and/or firing pattern variability, the ACC had a CC _r of 79.7% with a MAD of 4.7% compared to 76.6 and 6.9%, respectively, for the CC. For real signals, the ACC had a CC _r of 70.0% with a MAD of 6.3% compared to 64.9 and 6.4%, respectively, for the CC. The test results demonstrate that the ACC can manage both MUP shape variability as well as MU firing pattern variability. The ACC adapts to EMG signal characteristics to create dynamic data driven classification criteria so that the number of MUP assignments made reflects the signal complexity and the number of erroneous assignments is kept sufficiently low. The ability of the ACC to adjust to specific signal characteristics suggests that it can be successfully applied to a wide variety of EMG signals.     

A hybrid feature selection method for DNA microarray data

Gene expression profiles, which represent the state of a cell at a molecular level, have great potential as a medical diagnosis tool. In cancer classification, available training data sets are generally of a fairly small sample size compared to the number of genes involved. Along with training data limitations, this constitutes a challenge to certain classification methods. Feature (gene) selection can be used to successfully extract those genes that directly influence classification accuracy and to eliminate genes which have no influence on it. This significantly improves calculation performance and classification accuracy. In this paper, correlation-based feature selection (CFS) and the Taguchi-genetic algorithm (TGA) method were combined into a hybrid method, and the K-nearest neighbor (KNN) with the leave-one-out cross-validation (LOOCV) method served as a classifier for eleven classification profiles to calculate the classification accuracy. Experimental results show that the proposed method reduced redundant features effectively and achieved superior classification accuracy. The classification accuracy obtained by the proposed method was higher in ten out of the eleven gene expression data set test problems when compared to other classification methods from the literature.     

Probe selection algorithm for oligonucleotide array-based medium-resolution genotyping

Medium-resolution genotyping has the goal of distinguishing different subgroups instead of each element in a group. An oligonucleotide array provides an inexpensive, high-throughput method to identify differences in DNA sequence among individuals, which is fundamental for genotyping. As the cost and difficulty of designing and fabricating the oligonucleotide array dramatically increase with the number of probes used, it is therefore important to have a design with a minimum number of probes meeting the requirement of medium-resolution genotyping. The first algorithm for designing and selecting probes for oligonucleotide array-based medium-resolution typing is reported. The goal in deriving the algorithm was to select a minimum number of probes from a large probe set on the premise of minimum loss of resolution. The algorithm, which was based on entropy, conditional entropy and mutual information theory, was used to select the minimum number of probes from a large probe set. The algorithm was tested on a human leukocyte antigen (HLA) sequence data set. Thirty probes were selected from 390 probes for HLA-A, and 60 probes were selected from 767 probes for HLA-B. Although the number of probes was reduced by almost ten times, the distinguishability was reduced only a little, by 0.45% (from 99.90% to 99.45%) for HLA-A and 0.27% (from 99.84% to 99.57%) for HLA-B, respectively. This is a satisfactory and practical result.     

Survival analysis for high-dimensional,heterogeneous medical data: Exploring feature extraction as an alternative to feature selection

BackgroundIn clinical research, the primary interest is often the time until occurrence of an adverse event, i.e., survival analysis. Its application to electronic health records is challenging for two main reasons: (1) patient records are comprised of high-dimensional feature vectors, and (2) feature vectors are a mix of categorical and real-valued features, which implies varying statistical properties among features. To learn from high-dimensional data, researchers can choose from a wide range of methods in the fields of feature selection and feature extraction. Whereas feature selection is well studied, little work focused on utilizing feature extraction techniques for survival analysis.ResultsWe investigate how well feature extraction methods can deal with features having varying statistical properties. In particular, we consider multiview spectral embedding algorithms, which specifically have been developed for these situations. We propose to use random survival forests to accurately determine local neighborhood relations from right censored survival data. We evaluated 10 combinations of feature extraction methods and 6 survival models with and without intrinsic feature selection in the context of survival analysis on 3 clinical datasets. Our results demonstrate that for small sample sizes – less than 500 patients – models with built-in feature selection (Cox model with ℓ₁ penalty, random survival forest, and gradient boosted models) outperform feature extraction methods by a median margin of 6.3% in concordance index (inter-quartile range: [−1.2 % ;14.6 %]).ConclusionsIf the number of samples is insufficient, feature extraction methods are unable to reliably identify the underlying manifold, which makes them of limited use in these situations. For large sample sizes – in our experiments, 2500 samples or more – feature extraction methods perform as well as feature selection methods.     

Model for decomposition of the motor unit action potential 2 Analysis of simulated motor unit action potentials

A model for decomposition of the motor unit action potential (MUAP) which finds an optimal fit of a set of simulated single-fibre action potentials (SSFAPs) to the original MUAP is tested. The composition of SSFAPs which best produces the MUAP is assumed to carry information about the actual distribution of single-fibre action potentials generating the MUAP. Two methods are derived from the model. The first makes use of a fixed set of SSFAPs. In the second method, a gradually expanding set of SSFAPs is built, using a sequence of crosscorrelation, optimal reconstruction and subtraction. In the paper MUAPs are constructed under various well defined conditions. The MUAPs are decomposed by the two methods, and the results are compared with traditional MUAP parameters. Under these conditions, the model obtains parameters with closer biological connections compared with traditional measures.