Zheng Chongxun

White Blood Cell Image Segmentation Using On-line Trained Neural Network

This paper addresses a fast white blood cell (WBC) image segmentation scheme implemented by on-line trained neural network. A pre-selecting technique, based on mean shift algorithm and uniform sampling, is utilized as an initialization tool to largely reduce the training set while preserving the most valuable distribution information. Furthermore, particle swarm optimization (PSO) is adopted to train the network for a faster convergence and escaping from a local optimum. Experiment results show that under the compatible image segmentation accuracy, the training set and running time can be reduced significantly, compared with traditional training methods

A new method for identifying the life parameters via radar

It has been proved that the vital signs can be detected via radar. To better identify the life parameters such as respiration and heartbeat, a novel method combined with several signal processing techniques is presented. Firstly, to improve the signal-to-noise ratio (SNR) of the life signals, the signal accumulation technique by FFT is used. Then, to restrain the interferences produced by moving objects, a dual filtering algorithm (DFA) which is able to remove the interferences by tracing the interfering spectral peaks is proposed. Finally, the wavelet transform is applied to separate the heartbeat from the respiration signal. The method cannot only help to automatically detect the existence of human beings effectively, but also identifying the parameters like respiration, heartbeat, and body-moving signals significantly. Experimental results demonstrated that the method is very promising in identifying the life parameters via radar.It has been proved that the vital signs can be detected via radar. To better identify the life parameters such as respiration and heartbeat, a novel method combined with several signal processing techniques is presented. Firstly, to improve the signal-to-noise ratio (SNR) of the life signals, the signal accumulation technique by FFT is used. Then, to restrain the interferences produced by moving objects, a dual filtering algorithm (DFA) which is able to remove the interferences by tracing the interfering spectral peaks is proposed. Finally, the wavelet transform is applied to separate the heartbeat from the respiration signal. The method cannot only help to automatically detect the existence of human beings effectively, but also identifying the parameters like respiration, heartbeat, and body-moving signals significantly. Experimental results demonstrated that the method is very promising in identifying the life parameters via radar.

Mental tasks classification and their EEG structures analysis by using the growing hierarchical self-organizing map

The unsupervised method of growing hierarchical self-organizing map (GHSOM) was used to perform mental tasks classification. The GHSOM is an adaptive artificial neural network model with hierarchical architecture that is able to detect the hierarchical structure of data. The results indicate that GHSOM provides more detailed clustering information than SOM, and gives visual information about the separability of mental tasks in an intuitive way. The average classification accuracy across 130 task pairs by using GHSOM was up to 96.7%.

Adaboost for improving classification of left and right hand motor imagery tasks

The Adaboost classifier with Fisher discriminant analysis (FDA) as base learner is proposed to discriminate the left and right hand motor imagery tasks in this paper. Firstly, multichannel complexity and held power of EEG within 10-12Hz over two brain hemispheres are extracted as feature vectors, which characterize the brain features during hand motor imagination. Then with the Adaboost classifier, the satisfactory classification results on test data can be obtained. The maximum classification accuracy reaches to 89.29% and the maximum mutual information is 0.59bit. The primary results show that the Adaboost could effectively improve the classification accuracy of left and right hand motor imagery tasks, so that it has great potentials to mental tasks classification for BCI.

Non-Negative Matrix Factorizations of Spontaneous Electroencephalographic Signals for Classification

Non-negative matrix factorization (NMF) is an algorithm that is able to learn a parts-based representation. The paper proposes a new spontaneous EEG classification method for attention-related tasks. NMF was employed as feature extraction tool, which leads to more localized and sparse features than other two reference methods: power spectrum method and principal component analysis. With conventional back propagation neural network classifier, several experiments were carried out. It was showed that the NMF-ANN structure preserved the spatio-temporal characteristics of EEG signals

A new method to monitor depth of anesthesia based on the autocorrelation EEG signals

To have a safe, noninvasive, reliable and economic anesthetic depth indicator, the change of the a rhythm of electroencephalogram (EEG) signal on autocorrelation property during general intravenous anesthesia is investigated based on the effects of general anesthetics on the a rhythm of EEG in prefrontal cortex area. To synthesize the effects of correlated behavior, the contamination of muscle artifact is not removed from the EEG data. The autocorrelation analysis shows: 1) The EEG signals in prefrontal cortex area on autocorrelation are sensitive to different anesthesia depths during general anesthesia. The difference of autocorrelation trace from awareness to anesthesia is obvious. The change of autocorrelation trace is consistent with the anesthesia process; 2) The changes of autocorrelation in FP1-Cz channel and FP2-Cz channel with time are almost synchronous during general anesthesia. This means that 1-channel- recordings from prefrontal cortex are sufficient to monitor depth of anesthesia; 3) The value of autocorrelation in anesthesia fluctuates within a small range and is small than 20. This means that the value of autocorrelation is stable in anesthesia; 4) The differences of the range that the value of autocorrelation fluctuates in anesthesia present individual differences in a way. Being calculation simple, single channel and the transition of autocorrelation trace from awareness to anesthesia obvious, this technique may be ease to use, low running cost and can be applied in real time. Autocorrelation may provide a new method to monitor depth of anesthesia in clinic.

Analysis of Kolmogorov Complexity in Spontaneous EEG Signal and It's Application to Assessment of Mental Fatigue

In this paper, an attempt was made to evaluate mental fatigue using Kolmogorov complexity analysis of spontaneous EEG. EEG signals registered from Fp1 and Fp2 under different mental fatigue states were recorded and analyzed using Kolmogorov complexity (KC) to investigate whether mental fatigue could be assessed using this measure. The result indicates that the value of KC is strongly correlative with the mental fatigue state and the values of KC decrease with mental fatigue increasing. It may be possible to differentiate different mental fatigue level according to the value of KC. This method may be useful in further research and efforts to evaluate mental fatigue level objectively. It may also provide a basis for the study of effects of mental fatigue on central neural system.

Improved cable function to represent the excitation of peripheral nerves

The classical cable function has been used to represent the response of peripheral nerves stimulated by an external parallel electric field. This function cannot describe the excitation of peripheral nerves stimulated by a perpendicular electric field. In the present paper, responses of the Ranvier nodes to a transverse-field are thoroughly investigated by mathematic simulation. This simulation demonstrates that the excitation results from the net inward current driven by an external field. Based on a two-stage process, a novel model is introduced to describe peripheral nerves stimulated by a transverse-field, and the classical cable function is modified. Using this modified cable equation, the excitation threshold of peripheral nerves in a transverse field during MS is obtained. The modified cable equation can be used to represent the response of peripheral nerves by an arbitrary electric field.

Nonlinear entropy analysis for detecting focal cerebral ischemia

With an experimental model for SD (Sparague-Sawley) rat focal cerebral ischemia presented, the EEG signals from ischemic region and normal region are extracted. Then two entropy estimation methods, approximate entropy (ApEn) and power spectral entropy (PSE) are proposed to analyze EEG signals. Results show that the EEG signals from ischemic region and normal region can be distinguished by the ApEn and PSE analyses and they can be used to detect both the extent and location of focal ischemic cerebral injury.

Reconstructing phase space for nonlinear analysis of heart rate

In this paper we explore a new method for constructing phase space from time series of heart beats for using the nonlinear model of chaotic time series to analyze HRV signal. We explore whether it is unreasonable to directly reconstruct the phase space from the RR intervals. A new method is proposed to reconstruct the phase space from a derived time series using filtered delay coordinate map. The method is applied to clinical cases.