Brain-computer interface (BCI) provides a direct communicating and controlling approach between the brain and surrounding environment, which attracts a wide range of interest in the fields of brain science and artificial intelligence. It is a core to decode the electroencephalogram (EEG) feature in the BCI system. The decoding efficiency highly depends on the feature extraction and feature classification algorithms. In this paper, we first introduce the commonly-used EEG features in the BCI system. Then we introduce the basic classical algorithms and their advanced versions used in the BCI system. Finally, we present some new BCI algorithms proposed in recent years. We hope this paper can spark fresh thinking for the research and development of high-performance BCI system.
The feature extraction and feature selection are the important issues in pattern recognition. Based on the geometric algebra representation of vector, a new feature extraction method using blade coefficient of geometric algebra was proposed in this study. At the same time, an improved differential evolution (DE) feature selection method was proposed to solve the elevated high dimension issue. The simple linear discriminant analysis was used as the classifier. The result of the 10-fold cross-validation (10 CV) classification of public breast cancer biomedical dataset was more than 96% and proved superior to that of the original features and traditional feature extraction method.
Automated characterization of different vessel wall tissues including atherosclerotic plaques, branchings and stents from intravascular ultrasound (IVUS) gray-scale images was addressed. The texture features of each frame were firstly detected with local binary pattern (LBP), Haar-like and Gabor filter in the present study. Then, a Gentle Adaboost classifier was designed to classify tissue features. The methods were validated with clinically acquired image data. The manual characterization results obtained by experienced physicians were adopted as the golden standard to evaluate the accuracy. Results indicated that the recognition accuracy of lipidic plaques reached 94.54%, while classification precision of fibrous and calcified plaques reached 93.08%. High recognition accuracy can be reached up to branchings 93.20% and stents 93.50%, respectively.
Cough recognition provides important clinical information for the treatment of many respiratory diseases. A new Mel frequency cepstrum coefficient (MFCC) extracting method has been proposed on the basis of the distributional characteristics of cough spectrum. The whole frequency band was divided into several sub-bands, and the energy coefficient for each band was obtained by method of principle component analysis. Then non-uniform filter-bank in Mel frequency is designed to improve the extracting process of MFCC by distributing filters according to the spectrum energy coefficients. Cough recognition experiment using hidden Markov model was carried out, and the results showed that the proposed method could effectively improve the performance of cough recognition.
Heart sound is one of the common medical signals for diagnosing cardiovascular diseases. This paper studies the binary classification between normal or abnormal heart sounds, and proposes a heart sound classification algorithm based on the joint decision of extreme gradient boosting (XGBoost) and deep neural network, achieving a further improvement in feature extraction and model accuracy. First, the preprocessed heart sound recordings are segmented into four status, and five categories of features are extracted from the signals based on segmentation. The first four categories of features are sieved through recursive feature elimination, which is used as the input of the XGBoost classifier. The last category is the Mel-frequency cepstral coefficient (MFCC), which is used as the input of long short-term memory network (LSTM). Considering the imbalance of the data set, these two classifiers are both improved with weights. Finally, the heterogeneous integrated decision method is adopted to obtain the prediction. The algorithm was applied to the open heart sound database of the PhysioNet Computing in Cardiology(CINC) Challenge in 2016 on the PhysioNet website, to test the sensitivity, specificity, modified accuracy and F score. The results were 93%, 89.4%, 91.2% and 91.3% respectively. Compared with the results of machine learning, convolutional neural networks (CNN) and other methods used by other researchers, the accuracy and sensibility have been obviously improved, which proves that the method in this paper could effectively improve the accuracy of heart sound signal classification, and has great potential in the clinical auxiliary diagnosis application of some cardiovascular diseases.
Emotion plays an important role in people's cognition and communication. By analyzing electroencephalogram (EEG) signals to identify internal emotions and feedback emotional information in an active or passive way, affective brain-computer interactions can effectively promote human-computer interaction. This paper focuses on emotion recognition using EEG. We systematically evaluate the performance of state-of-the-art feature extraction and classification methods with a public-available dataset for emotion analysis using physiological signals (DEAP). The common random split method will lead to high correlation between training and testing samples. Thus, we use block-wise K fold cross validation. Moreover, we compare the accuracy of emotion recognition with different time window length. The experimental results indicate that 4 s time window is appropriate for sampling. Filter-bank long short-term memory networks (FBLSTM) using differential entropy features as input was proposed. The average accuracy of low and high in valance dimension, arousal dimension and combination of the four in valance-arousal plane is 78.8%, 78.4% and 70.3%, respectively. These results demonstrate the advantage of our emotion recognition model over the current studies in terms of classification accuracy. Our model might provide a novel method for emotion recognition in affective brain-computer interactions.
The clinical manifestations of patients with schizophrenia and patients with depression not only have a certain similarity, but also change with the patient's mood, and thus lead to misdiagnosis in clinical diagnosis. Electroencephalogram (EEG) analysis provides an important reference and objective basis for accurate differentiation and diagnosis between patients with schizophrenia and patients with depression. In order to solve the problem of misdiagnosis between patients with schizophrenia and patients with depression, and to improve the accuracy of the classification and diagnosis of these two diseases, in this study we extracted the resting-state EEG features from 100 patients with depression and 100 patients with schizophrenia, including information entropy, sample entropy and approximate entropy, statistical properties feature and relative power spectral density (rPSD) of each EEG rhythm (δ, θ, α, β). Then feature vectors were formed to classify these two types of patients using the support vector machine (SVM) and the naive Bayes (NB) classifier. Experimental results indicate that: ① The rPSD feature vector P performs the best in classification, achieving an average accuracy of 84.2% and a highest accuracy of 86.3%; ② The accuracy of SVM is obviously better than that of NB; ③ For the rPSD of each rhythm, the β rhythm performs the best with the highest accuracy of 76%; ④ Electrodes with large feature weight are mainly concentrated in the frontal lobe and parietal lobe. The results of this study indicate that the rPSD feature vector P in conjunction with SVM can effectively distinguish depression and schizophrenia, and can also play an auxiliary role in the relevant clinical diagnosis.
Feature extraction is a very crucial step in P300-based brain-computer interface (BCI) and independent component analysis (ICA) is a suitable P300 feature extraction method. But at present the convergence performance of the general ICA iteration methods are not very satisfactory. In this paper, a method based on quantum particle swarm optimizer (QPSO) algorithm and ICA technique is put forward for P300 extraction. In this method, quantum computing is used to impel ICA iteration to globally converge faster. It achieved the purpose of extracting P300 rapidly and efficiently. The method was tested on two public datasets of BCI Competition Ⅱ and Ⅲ, and a simple linear classifier was employed to classify the extracted P300 features. The recognition accuracy reached 94.4% with 15 times averaged. The results showed that the proposed method could extract P300 rapidly and the extraction effect did not reduce. It provides an experimental basis for further study of real-time BCI system.
Automatic classification of different types of cough plays an important role in clinical. In the previous research of cough classification or cough recognition, traditional Mel frequency cepstrum coefficients (MFCC) which extracts feature mainly from low frequency band is usually used as feature expression. In this paper, by analyzing the distributions of spectral energy of dry/wet cough, it is found that spectral difference of two types of cough exits mainly in middle frequency band and high frequency band. To better reflect the spectral difference of dry cough and wet cough, an improved method of extracting reverse MFCC is proposed. In this method, reverse Mel filter-bank in which filters are allocated in reverse Mel scale is adopted and is improved by placing filters only in the frequency band with high spectral energy. As a result, features are mainly extracted from the frequency band where two types of cough show both high spectral energy and distinguished difference. Detailed process of accessing improved reverse MFCC was introduced and hidden Markov models trained by 60 dry cough and 60 wet cough were used as cough classification model. Classification experiment results for 120 dry cough and 85 wet cough showed that, compared to traditional MFCC, better classification performance was achieved by the proposed method and the total classification accuracy was raised from 89.76% to 93.66%.
Due to the sparsity of brain encoding, the neural ensemble signals recorded by microelectrode arrays contain a lot of noise and redundant information, which could reduce the stability and precision of decoding of motion intent. To solve this problem, we proposed a decoding method based on partial least squares (PLS) feature extraction in our study. Firstly, we extracted the features of spike signals using the PLS, and then classified them with support vector machine (SVM) classifier, and decoded them for motion intent. In this study, we decoded neural ensemble signals based on plus-maze test. The results have shown that the proposed method had a better stability and higher decoding accuracy, due to the PLS combined with classification model which overcame the shortcoming of PLS regression that was easily affected by accumulated effect of noise. Meanwhile, the PLS method extracted fewer features with more useful information in comparison with common feature extraction method. The decoding accuracy of real data sets were 93.59%, 84.00% and 83.59%, respectively.