Hidden Markov models for human interactive behavior / 生物医学工程学杂志

In order to meet the requirements in the cooperation and competition experiments for an individual patient in clinical application, two human interactive behavior key-press models based on hidden Markov model (HMM) were proposed. To validate the cooperative and competitive models, a verification experimental task was designed and the data were collected. The correlation of the score and subjects' participation level has been used to analyze the reasonability verification. Behavior verification was conducted by comparing the statistical difference in response time for subjects between human-human and human-computer experiment. In order to verify the physiological validity of the models, we have utilized the coherence analysis to analyze the deep information of prefrontal brain area. Reasonability verification shows that the correlation coefficient for the training data and the testing data is 0.883 1 and 0.578 6 respectively based on cooperation model, and 0.813 1 and 0.617 8 respectively based on the competition model. The behavioral verification result shows that the cooperation and competition models have an accuracy of 71.43% respectively. The results of physiological validity show that the deep information of prefrontal brain area could been extracted based on the cooperation and competition models, and reveal the consistency of coherence between the double key-press cooperative and competitive experiments, respectively. Above all, the high consistency is obtained between the cooperatio/competition model and the double key-press experiment by the behavioral and physiological evaluation results. Consequently, the cooperation and competition models could be applied to clinical trials.

Feature extraction and classification of EEG for mental tasks based on wavelet packet analysis / 生物医学工程学杂志

This paper explores the use of wavelet packet analysis to extract features from spontaneous electroencephalogram (EEG) during three different mental tasks. Artifact-free EEG segments are transformed to multi-scale representations by dyadic wavelet packet decomposition channel by channel. Their feature vectors formed by energy values of different sub-spaces EEG components are used as inputs of a radial basis function network to test the classification accuracies of three task pairs. The results indicate that the classification accuracies of the wavelet packet analysis method are significantly better than those of autoregressive model method. Wavelet packet analysis would be a promising method to extract features from EEG signals.

Application of support vector machines to classification of blood cells / 生物医学工程学杂志

The support vector machine (SVM) is a new learning technique based on the statistical learning theory. It was originally developed for two-class classification. In this paper, the SVM approach is extended to multi-class classification problems, a hierarchical SVM is applied to classify blood cells in different maturation stages from bone marrow. Based on stepwise decomposition, a hierarchical clustering method is presented to construct the architecture of the hierarchical (tree-like) SVM, then the optimal control parameters of SVM are determined by some criterion for each discriminant step. To verify the performances of classifiers, the SVM method is compared with three classical classifiers using 3-fold cross validation. The preliminary results indicate that the proposed method avoids the curse of dimensionality and has greater generalization. Thus, the method can improve the classification correctness for blood cells from bone marrow.

Study on automatic segmentation of color images applied to blood cells / 生物医学工程学杂志

A hybrid segmentation algorithm is proposed for automatic segmentation of blood cell images based on adaptive multi-scale thresholding and seeded region growing techniques. Firstly, an adaptive and scale space filter (ASSF) is applied to image histogram and a scale space image is built. According to the properties of the scale space image, proper thresholds can be obtained to separate the nucleus from the original image and the white blood cells are located. Secondly, the local color similarity and global morphological criteria constrain seeded region growing in order to finish the segmentation of the cytoplasm. The detection accuracy of white blood cell is 98% and the segmentation accuracy based on the subjective evaluation is 93%. Test shows that this algorithm is effective for automatic segmentation of white blood cells.