Chen-Wen Yen

Improving the generalization performance of RBF neural networks using a linear regression technique

In this paper we present a method for improving the generalization performance of a radial basis function (RBF) neural network. The method uses a statistical linear regression technique which is based on the orthogonal least squares (OLS) algorithm. We first discuss a modified way to determine the center and width of the hidden layer neurons. Then, substituting a QR algorithm for the traditional Gram-Schmidt algorithm, we find the connected weight of the hidden layer neurons. Cross-validation is utilized to determine the stop training criterion. The generalization performance of the network is further improved using a bootstrap technique. Finally, the solution method is used to solve a simulation and a real problem. The results demonstrate the improved generalization performance of our algorithm over the existing methods.

Characterizing gait asymmetry via frequency sub-band components of the ground reaction force

Abstract This study introduces gait asymmetry measures by comparing the ground reaction force (GRF) features of the left and right limbs. The proposed features were obtained by decomposing the GRF into components of different frequency sub-bands via the wavelet transform. The correlation coefficients between the right and left limb GRF components of the same frequency sub-band were used to characterize the degree of bilateral symmetry. The asymmetry measures were then obtained by subtracting these coefficients from one. To demonstrate the effectiveness of these asymmetry measures, the proposed measures were applied to differentiate the walking patterns of Parkinsons patients and healthy subjects. The results of the statistical analyses found that the patient group has a higher degree of gait asymmetry. By comparing these results with those obtained by conventional asymmetry measures, it was found that the proposed approach can more effectively distinguish the differences between the tested Parkinsons disease patients and the healthy control subjects.

A vector quantization method for nearest neighbor classifier design

This paper proposes a nearest neighbor classifier design method based on vector quantization (VQ). By investigating the error distribution pattern of the training set, the VQ technique is applied to generate prototypes incrementally until the desired classification result is reached. Experimental results demonstrate the effectiveness of the method.

Detecting Slow Wave Sleep Using a Single EEG Signal Channel

BACKGROUND In addition to the cost and complexity of processing multiple signal channels, manual sleep staging is also tedious, time consuming, and error-prone. The aim of this paper is to propose an automatic slow wave sleep (SWS) detection method that uses only one channel of the electroencephalography (EEG) signal. NEW METHOD The proposed approach distinguishes itself from previous automatic sleep staging methods by using three specially designed feature groups. The first feature group characterizes the waveform pattern of the EEG signal. The remaining two feature groups are developed to resolve the difficulties caused by interpersonal EEG signal differences. RESULTS AND COMPARISON WITH EXISTING METHODS The proposed approach was tested with 1,003 subjects, and the SWS detection results show kappa coefficient at 0.66, an accuracy level of 0.973, a sensitivity score of 0.644 and a positive predictive value of 0.709. By excluding sleep apnea patients and persons whose age is older than 55, the SWS detection results improved to kappa coefficient, 0.76; accuracy, 0.963; sensitivity, 0.758; and positive predictive value, 0.812. CONCLUSIONS With newly developed signal features, this study proposed and tested a single-channel EEG-based SWS detection method. The effectiveness of the proposed approach was demonstrated by applying it to detect the SWS of 1003 subjects. Our test results show that a low SWS ratio and sleep apnea can degrade the performance of SWS detection. The results also show that a large and accurately staged sleep dataset is of great importance when developing automatic sleep staging methods.

Basal sympathetic predominance in periodic limb movements in sleep with obstructive sleep apnea

Because the impact of periodic limb movements in sleep (PLMS) is controversial, no consensus has been reached on the therapeutic strategy for PLMS in obstructive sleep apnea (OSA). To verify the hypothesis that PLMS is related to a negative impact on the cardiovascular system in OSA patients, this study investigated the basal autonomic regulation by heart rate variability (HRV) analysis. Sixty patients with mild‐to‐moderate OSA who underwent polysomnography (PSG) and completed sleep questionnaires were analysed retrospectively and divided into the PLMS group (n = 30) and the non‐PLMS group (n = 30). Epochs without any sleep events or continuous effects were evaluated using HRV analysis. No significant difference was observed in the demographic data, PSG parameters or sleep questionnaires between the PLMS and non‐PLMS groups, except for age. Patients in the PLMS group had significantly lower normalized high frequency (n‐HF), high frequency (HF), square root of the mean of the sum of the squares of difference between adjacent NN intervals (RMSSD) and standard deviation of all normal to normal intervals index (SDNN‐I), but had a higher normalized low frequency (n‐LF) and LF/HF ratio. There was no significant difference in the Epworth Sleepiness Scale, the Pittsburgh Sleep Quality Index, the Short‐Form 36 and the Hospital Anxiety and Depression Scale between the two groups. After adjustment for confounding variables, PLMS remained an independent predictor of n‐LF (β = 0.0901, P = 0.0081), LF/HF ratio (β = 0.5351, P = 0.0361), RMSSD (β = −20.1620, P = 0.0455) and n‐HF (β = −0.0886, P = 0.0134). In conclusion, PLMS is related independently to basal sympathetic predominance and has a potentially negative impact on the cardiovascular system of OSA patients.

DETERMINING THE ASSOCIATION BETWEEN DERMATOGLYPHICS AND SCHIZOPHRENIA BY USING FINGERPRINT ASYMMETRY MEASURES

Early detection and intervention strategies for schizophrenia are receiving increasingly more attention. Dermatoglyphic patterns, such as the degree of asymmetry of the fingerprints, have been hypothesized to be indirect measures for early abnormal developmental processes that can lead to later psychiatric disorders such as schizophrenia. However, previous results have been inconsistent in trying to establish the association between dermatoglyphics and schizophrenia. The goal of this work is to try to resolve this problem by borrowing well-developed techniques from the field of fingerprint matching. Two dermatoglyphic asymmetry measures are proposed that draw on the orientation field of homologous fingers. To test the capability of these measures, fingerprint images were acquired digitally from 40 schizophrenic patients and 51 normal individuals. Based on these images, no statistically significant association between conventional dermatoglyphic asymmetry measures and schizophrenia was found. In contrast, the sample means of the proposed measures consistently identified the patient group as having a higher degree of asymmetry than the control group. These results suggest that the proposed measures are promising for detecting the dermatoglyphic patterns that can differentiate the patient and control groups.

A neural network-based diagnostic method for solitary pulmonary nodules

Several computer-aided diagnostic (CAD) methods for solitary pulmonary nodules (SPNs) have been proposed, which can be divided into two major categories: (1) the morphometric CT method, depending on high-resolution morphometric characteristics from single CT scan and (2) the perfusion CT method, depending on properties of the post-contrast enhancement dynamics obtained from repeated CT scans at predefined time points. The goal of this work is to introduce a neural network-based CAD method of lung nodule diagnosis by combining morphometry and perfusion characteristics by perfusion CT. Compared with previous methods, the proposed approach has the following distinctive features. Firstly, this work develops a very efficient semi-automatic procedure to segment entire nodules. Secondly, reliable nodule classification can be achieved by using only two time-point perfusion CT feature measures (precontrast and 90s). This greatly reduces the amount of radiation exposure to patients and the data processing time. The effectiveness of the proposed approach is compared with those of several previously developed CAD methods.

A comparative study for 2D and 3D computer-aided diagnosis methods for solitary pulmonary nodules

Many computer-aided diagnosis (CAD) methods, including 2D and 3D approaches, have been proposed for solitary pulmonary nodules (SPNs). However, the detection and diagnosis of SPNs remain challenging in many clinical circumstances. One goal of this work is to investigate the relative diagnostic accuracy of 2D and 3D methods. An additional goal is to develop a two-stage approach that combines the simplicity of 2D and the accuracy of 3D methods. The experimental results show statistically significant differences between the diagnostic accuracy of 2D and 3D methods. The results also show that with a very minor drop in diagnostic performance the two-stage approach can significantly reduce the number of nodules needed to be processed by the 3D method, streamlining the computational demand.

Gait verification using knee acceleration signals

A novel gait recognition method for biometric applications is proposed. The approach has the following distinct features. First, gait patterns are determined via knee acceleration signals, circumventing difficulties associated with conventional vision-based gait recognition methods. Second, an automatic procedure to extract gait features from acceleration signals is developed that employs a multiple-template classification method. Consequently, the proposed approach can adjust the sensitivity and specificity of the gait recognition system with great flexibility. Experimental results from 35 subjects demonstrate the potential of the approach for successful recognition. By setting sensitivity to be 0.95 and 0.90, the resulting specificity ranges from 1 to 0.783 and 1.00 to 0.945, respectively.

One-Class-at-a-Time Removal Sequence Planning Method for Multiclass Classification Problems

Using dynamic programming, this work develops a one-class-at-a-time removal sequence planning method to decompose a multiclass classification problem into a series of two-class problems. Compared with previous decomposition methods, the approach has the following distinct features. First, under the one-class-at-a-time framework, the approach guarantees the optimality of the decomposition. Second, for a K-class problem, the number of binary classifiers required by the method is only K-1. Third, to achieve higher classification accuracy, the approach can easily be adapted to form a committee machine. A drawback of the approach is that its computational burden increases rapidly with the number of classes. To resolve this difficulty, a partial decomposition technique is introduced that reduces the computational cost by generating a suboptimal solution. Experimental results demonstrate that the proposed approach consistently outperforms two conventional decomposition methods