Adaptive motor unit potential train validation using MUP shape information

Cross Comparison of Motor Unit Potential Features Used in EMG Signal Decomposition

Feature extraction is an important step of resolving an electromyographic (EMG) signal into its component motor unit potential trains, commonly known as EMG decomposition. Until now, different features have been used to represent motor unit potentials (MUPs) and improve decomposition processing time and accuracy, but a major limitation is that no systematic comparison of these features exists. In an EMG decomposition system, like any pattern recognition system, the features used for representing MUPs play an important role in the overall performance of the system. A cross comparison of the feature extraction methods used in EMG signal decomposition can assist in choosing the best features for representing MUPs and ultimately may improve EMG decomposition results. This paper presents a survey and cross comparison of these feature extraction methods. Decomposability index, classification accuracy of a -nearest neighbors classifier, and class-feature mutual information were employed for evaluating the discriminative power of various feature extraction techniques commonly used in the literature including time domain, morphological, frequency domain, and discrete wavelets. In terms of data, 45 simulated and 82 real EMG signals were used. Results showed that among time domain features, the first derivative of time samples exhibit the best separability. For morphological features, slope analysis provided the most discriminative power. Discrete Fourier transform coefficients offered the best separability among frequency domain features. However, neither morphological nor frequency domain techniques outperformed time domain features. The detail 4 coefficients in a discrete wavelets decomposition exceeded in evaluation measures when compared with other feature extraction techniques. Using principal component analysis slightly improved the results, but it is time consuming. Overall, considering computation time and discriminative ability, the first derivative of time samples might be efficient in representing MUPs in EMG decomposition and there is no need for sophisticated feature extraction methods.

Automatic segmentation of vertebrae in 3D CT images using adaptive fast 3D pulse coupled neural networks

Two systems are presented for segmentation of vertebrae in a 3D computed tomography (CT) image. The first method extracts seven features from each voxel and uses a multi-layer perceptron neural network (MLPNN) to classify the voxel as vertebrae or background. In the second method, the segmentation is completed in two steps: first, a newly developed adaptive pulse coupled neural network (APCNN) directly applied to a given image segments vertebrae, then the result is refined using a median filter. In the developed APCNN, the values for the user-defined parameters of the pulse coupled neural networks (PCNN) are adaptively adjusted for each image individually, instead of using one value for all images as in conventional PCNN. The performance of both systems in terms of Dice index (DI) was evaluated and compared against the state-of-the-art segmentation methods using seventeen clinical and standard CT images. Overall, both systems demonstrated statistically similar and promising performance with average DI > 95%. Compared to existing PCNN-based segmentation algorithms, the accuracy of the proposed APCNN improved by 29.3% on average. The developed APCNN-based system is more accurate than MLPNN-based system and existing PCNN-based algorithms in segmentation of vertebrae with blurred and weak boundaries and in the images contaminated by salt- and- pepper noise. In terms of computation time, the APCNN-based system is 16 times faster than the MLPNN-based system. Consequently, the presented APCNN-based algorithm is both accurate and fast and could be used in clinical environment for segmentation of vertebrae in 3D CT images.

A SYSTEM FOR ACCURATELY PREDICTING THE RISK OF MYOCARDIAL INFARCTION USING PCG, ECG AND CLINICAL FEATURES

Myocardial infarction (MI) also known as heart attack is one of the prevalence cardiovascular diseases. MI that is due to the blockade in the coronary artery is caused by the lack of blood supply (ischemia) to heart tissue. Determining the risk of MI and hospitalizing the victim immediately can prolong patient’s life and enhance the quality of living through appropriate treatment. To make this decision more accurate, in this study, a decision support system is proposed to classify patients with hard chest pain (sign of MI) into high and low risk groups. Such a system can also assist in managing the limitation of bed in the care units such as cardiac care unit by deciding on admitting a subject with a hard chest pain whom refers to a hospital or not. Despite several efforts in this issue, the so far published results demonstrated that distinguishing these patients using just electrocardiogram (ECG) features is not promising. In addition, these methods did not focus on classifying the patients with high and low risks of MI. In this regard, auxiliary features from phonocardiogram (PCG) signals and clinical data were elicited to create a discriminative feature set and ultimately improve the performance of the decision making system. In this research, ECG (from 12 leads), PCG signal and clinical data were acquired from 83 patients two times (morning and evening) in the first day. Since the number of elicited features from the raw data of each patient is high, the irrelevant and non-discriminative features were eliminated by sequential forward selection. The selected features were applied to [Formula: see text]-nearest neighbor classifier resulted in 98.0% sensitivity, 100% specificity and 99.0% accuracy over the patients. The results illustrate that neither clinical data nor ECG features nor PCG features are lonely enough for estimating the risk of MI. Employing features from different modalities can improve the performance such that the developed multimodal-based system overperformed single modal-based systems. The obtained results are promising and suggest that using this system might be useful as a means for altering the risk of MI in patients.

Mechanistic Hierarchical Gaussian Processes

The statistics literature on functional data analysis focuses primarily on flexible black-box approaches, which are designed to allow individual curves to have essentially any shape while characterizing variability. Such methods typically cannot incorporate mechanistic information, which is commonly expressed in terms of differential equations. Motivated by studies of muscle activation, we propose a nonparametric Bayesian approach that takes into account mechanistic understanding of muscle physiology. A novel class of hierarchical Gaussian processes is defined that favors curves consistent with differential equations defined on motor, damper, spring systems. A Gibbs sampler is proposed to sample from the posterior distribution and applied to a study of rats exposed to non-injurious muscle activation protocols. Although motivated by muscle force data, a parallel approach can be used to include mechanistic information in broad functional data analysis applications.