Investigating the contribution of distance-based features to automatic sleep stage classification.

OBJECTIVE: In this paper, the contribution of distance-based features to automatic sleep stage classification is investigated. The potency of these features is analyzed individually and in combination with 48 conventionally used features. METHODS: The distance-based set consists of 32 features extracted by calculating Itakura, Itakura-Saito and COSH distances of autoregressive and spectral coefficients of Electrocardiography (EEG) (C3-A2), Left EOG, Chin EMG and ECG signals. All the evaluations are performed on three feature sets: distance-based, conventional and total (combined distance based and conventional). Six ranking methods were used to find the top features with the highest discrimination ability in each set. The ranked feature lists were evaluated using k-Nearest Neighbor (kNN), Artificial Neural Network (ANN), and Decision-tree-based multi-SVM (DSVM) classifiers for five sleep stages including Wake, REM, N1, N2 and N3. Furthermore, the ability of distance-based and conventional features to discriminate between each pair of sleep stages was evaluated using t-test, a hypothesis testing method. RESULTS: Distance-based features occupied 25% of top-ranked features. Simulation results showed that using distance-based features together with conventional features can lead to an enhancement of accuracy. The best classification accuracy (85.5%) was achieved by DSVM classifier and 13 features selected by mRMR-MID and normalized with Min-Max method for total feature set, where two of them were from the distance-based feature set. The t-test results show that distance-based features outperform conventional features in discriminating between N1 and REM stages that is usually a challenge for classification systems. CONCLUSION: Distance-based features have a positive contribution to sleep stage classification, including enhancement of accuracy and better REM-N1 discrimination ability. SIGNIFICANCE: The main motivation for this work was to evaluate new features to characterize each sleep stage in such a way that extracted features were more powerful than conventional features, to distinguish sleep stages from each other, and to improve classifiers accuracy.

Feature ranking and rank aggregation for automatic sleep stage classification: a comparative study.

BACKGROUND: Nowadays, sleep quality is one of the most important measures of healthy life, especially considering the huge number of sleep-related disorders. Identifying sleep stages using polysomnographic (PSG) signals is the traditional way of assessing sleep quality. However, the manual process of sleep stage classification is time-consuming, subjective and costly. Therefore, in order to improve the accuracy and efficiency of the sleep stage classification, researchers have been trying to develop automatic classification algorithms. Automatic sleep stage classification mainly consists of three steps: pre-processing, feature extraction and classification. Since classification accuracy is deeply affected by the extracted features, a poor feature vector will adversely affect the classifier and eventually lead to low classification accuracy. Therefore, special attention should be given to the feature extraction and selection process. METHODS: In this paper the performance of seven feature selection methods, as well as two feature rank aggregation methods, were compared. Pz-Oz EEG, horizontal EOG and submental chin EMG recordings of 22 healthy males and females were used. A comprehensive feature set including 49 features was extracted from these recordings. The extracted features are among the most common and effective features used in sleep stage classification from temporal, spectral, entropy-based and nonlinear categories. The feature selection methods were evaluated and compared using three criteria: classification accuracy, stability, and similarity. RESULTS: Simulation results show that MRMR-MID achieves the highest classification performance while Fisher method provides the most stable ranking. In our simulations, the performance of the aggregation methods was in the average level, although they are known to generate more stable results and better accuracy. CONCLUSIONS: The Borda and RRA rank aggregation methods could not outperform significantly the conventional feature ranking methods. Among conventional methods, some of them slightly performed better than others, although the choice of a suitable technique is dependent on the computational complexity and accuracy requirements of the user.

A multichannel time-frequency and multi-wavelet toolbox for uterine electromyography processing and visualisation.

The uterine electromyogram, also called electrohysterogram (EHG), is an electrical signal generated by the uterine contractile activity. The EHG has been considered a promising biomarker for labour and preterm labour prediction, for which there is a demand for accurate estimation methods. Preterm labour is a significant public health concern and one of the major causes of neonatal mortality and morbidity [1]. Given the non-stationary properties of the EHG signal, time-frequency domain analysis can be used. For real life signals it is not generally possible to determine a priori the suitable quadratic time-frequency kernel or the appropriate wavelet family and relative parameters, regarding, for instance, the adequate detection of the signal frequency variation in time. There has been a lack of a comprehensive software tool for the selection of the appropriate time frequency representation of a multichannel EHG signal and extraction of relevant spectral and temporal information. The presented toolbox (Uterine Explorer) has been specifically designed for the EHG analysis and exploration in view of the characterisation of its components. The starting point is the multichannel scalogram or spectrogram representation from which frequency and time marginals, instantaneous frequency and bandwidth are obtained as EHG features. From this point the detected components undergo parametric and non-parametric spectral estimation and wavelet packet analysis. Intrauterine pressure estimation (IUP) is obtained using the Teager, RMS, wavelet marginal and Hilbert operators over the EHG. This toolbox has been tested to build up a dictionary of 288 EHG components [2], useful for research in preterm labour prediction.