Cyclostationary analysis of uterine EMG measurements for the prediction of preterm birth.

Preterm birth (gestational age < 37 weeks) is a public health concern that causes fetal and maternal mortality and morbidity. When this condition is detected early, suitable treatment can be prescribed to delay labour. Uterine electromyography (uEMG) has gained a lot of attention for detecting preterm births in advance. However, analyzing uEMG is challenging due to the complexities associated with inter and intra-subject variations. This work aims to investigate the applicability of cyclostationary characteristics in uEMG signals for predicting premature delivery. The signals under term and preterm situations are considered from two online datasets. Preprocessing is carried out using a Butterworth bandpass filter, and spectral correlation density function is adapted using fast Fourier transform-based accumulation method (FAM) to compute the cyclostationary variations. The cyclic frequency spectral density (CFSD) and degree of cyclostationarity (DCS) are quantified to assess the existence of cyclostationarity. Features namely, maximum cyclic frequency, bandwidth, mean cyclic frequency (MNCF), and median cyclic frequency (MDCF) are extracted from the cyclostationary spectrum and analyzed statistically. uEMG signals exhibit cyclostationarity property, and these variations are found to distinguish preterm from term conditions. All the four extracted features are noted to decrease from term to preterm conditions. The results indicate that the cyclostationary nature of the signals can provide better characterization of uterine muscle contractions and could be helpful in detecting preterm birth. The proposed method appears to aid in detecting preterm birth, as analysis of uterine contractions under preterm conditions is imperative for timely medical intervention.

Empirical Mode Decomposition Based Measures for Investigating the Progression of Pregnancy from Uterine EMG.

The objective of this study is to analyze the uterine electromyography (uEMG) signals to study the progression of pregnancy under term condition (gestational age > 36 weeks) using EMD-based time-frequency features. uEMG signals are obtained from the multiple public datasets during two conditions, namely T1 (acquired < 26 gestational weeks) and T2 (acquired ≥ 26 gestational weeks). The considered signals are preprocessed. Empirical mode decomposition is applied to decompose the signals and time-frequency features, such as median frequency (MDF), mean frequency (MNF), peak frequency and peak magnitude, are extracted from each intrinsic mode functions and statistically analyzed. The results depict that the obtained time-frequency features are able to distinguish between T1 and T2 conditions. The extracted features, namely MNF and MDF, are observed to decrease from T1 to T2 conditions. These features are found to have higher effect size, confirming the better differentiation between T1 and T2 conditions. It appears that EMD-based time-frequency features can aid in studying the evolving changes in uterine contractions towards labor.

Automated Detection of Seizure Types from the Higher-Order Moments of Maximal Overlap Wavelet Distribution.

In this work, an attempt has been made to develop an automated system for detecting electroclinical seizures such as tonic-clonic seizures, complex partial seizures, and electrographic seizures (EGSZ) using higher-order moments of scalp electroencephalography (EEG). The scalp EEGs of the publicly available Temple University database are utilized in this study. The higher-order moments, namely skewness and kurtosis, are extracted from the temporal, spectral, and maximal overlap wavelet distributions of EEG. The features are computed from overlapping and non-overlapping moving windowing functions. The results show that the wavelet and spectral skewness of EEG is higher in EGSZ than in other types. All the extracted features are found to have significant differences (p < 0.05), except for temporal kurtosis and skewness. A support vector machine with a radial basis kernel designed using maximal overlap wavelet skewness yields a maximum accuracy of 87%. In order to improve the performance, the Bayesian optimization technique is utilized to determine the suitable kernel parameters. The optimized model achieves the highest accuracy of 96% and an MCC of 91% in three-class classification. The study is found to be promising, and it could facilitate the rapid identification process of life-threatening seizures.

Analysis of Facial Electromyography Signals Using Linear and Non-Linear Features for Human-Machine Interface.

In this work, an attempt has been made to analyze the facial electromyography (facial EMG) signals using linear and non-linear features for the human-machine interface. Facial EMG signals are obtained from the publicly available, widely used DEAP dataset. Thirty-two healthy subjects volunteered for the establishment of this dataset. The signals of one positive emotion (joy) and one negative emotion (sadness) obtained from the dataset are used for this study. The signals are segmented into 12 epochs of 5 seconds each. Features such as sample entropy and root mean square (RMS) are extracted from each epoch for analysis. The results indicate that facial EMG signals exhibit distinct variations in each emotional stimulus. The statistical test performed indicates statistical significance (p<0.05) in various epochs. It appears that this method of analysis could be used for developing human-machine interfaces, especially for patients with severe motor disabilities such as people with tetraplegia.

Detection of Tonic-Clonic Seizures using Wavelet Entropy of Scalp EEG.

Epilepsy is the most common chronic neurologic disorder characterized by the recurrence of unprovoked seizures. These seizures are paroxysmal events that result from abnormal neuronal discharges and are categorized into various types based on the clinical manifestations and localization. Tonic-Clonic seizures (TCSZ) may lead to injuries, and constitute the major risk factor for sudden unexpected death in epilepsy (SUDEP), especially in unattended patients. Therapeutic decisions and clinical trials rely on Video EEG which is not practical outside of clinical setting. In this study, wavelet entropy of scalp EEG signals are utilized to discriminate the seizures with and without clinical manifestations. The scalp EEG records from the publically available Temple University Hospital (TUH) dataset are considered for this work. A sevenlevel, fourth order Daubechies (db4) wavelet is utilized for the decomposition of first four seconds of scalp EEG during seizures. The entropy is extracted from the resultant coefficients and are used to develop SVM based models. Most of the extracted features found to have significant differences (p<0.05). The results show that polynomial SVM model achieves an accuracy of 95.5%, positive predictive value (PPV) of 99.4%, negative predictive value (NPV) of 91.57% and F-Score of 95.9%. Therefore, the proposed approach could be a support in detecting life-threatening seizures.

Emotion Recognition Using Spectral Feature from Facial Electromygraphy Signals for Human-Machine Interface.

Recognition of the emotions demonstrated by human beings plays a crucial role in healthcare and human-machine interface. This paper reports an attempt to classify emotions using a spectral feature from facial electromyography (facial EMG) signals in the valence affective dimension. For this purpose, the facial EMG signals are obtained from the DEAP dataset. The signals are subjected to Short-Time Fourier Transform, and the peak frequency values are extracted from the signal in intervals of one second. Support vector machine (SVM) classifier is used for the classification of the features extracted. The extracted feature can classify the signals in the valence dimension with an accuracy of 61.37%. The proposed feature could be used as an added feature for emotion recognition, and this method of analysis could be extended to myoelectric control applications.

Analysis of Frequency Bands of Uterine Electromyography Signals for the Detection of Preterm Birth.

In this work, an attempt has been made to analyze the influence of the frequencies bands in uterine electromyography (uEMG) signals on the detection of preterm birth. The signals recorded from the women's abdomen during pregnancy are considered in this study. The signals are subjected to preprocessing using digital bandpass Butterworth filter and decomposed into different frequency bands namely, 0.3-1.0 Hz (F1), 1.0-2.0 Hz (F2) and 2.0-3.0Hz (F3). Spectral features namely, peak magnitude, peak frequency, mean frequency and median frequency are extracted from the power spectrum. Classification models namely, k-nearest neighbor, support vector machine and random forest are employed to distinguish the term and preterm conditions. The results show that the features extracted from these frequency bands are able to differentiate term and preterm condition. Particularly, the frequency band F3 performs better than other frequency bands. The features associated with these frequencies along with random forest classification model achieves a maximum accuracy of 75.2%. Thus, these measures could be used to accurately detect the preterm birth well in advance.

Automated detection of subthalamic nucleus in deep brain stimulation surgery for Parkinson's disease using microelectrode recordings and wavelet packet features.

BACKGROUND: Deep brain stimulation (DBS) to the subthalamic nucleus (STN) is an effective neurosurgery that overcomes the motor system alternations of patients with advanced Parkinson's disease. The most challenging aspect of DBS surgery is the accurate identification of STN and its borders. In general, it is performed manually by a neurophysiologist using the microelectrode recordings (MERs). This process is subjective, and tedious and further, interpretation of MERs is difficult because of its inherent nonstationary variations. NEW METHODS: In this work, the wavelet-packet based features are proposed to automatically localize the STN and its subcortical structures using microelectrode recorded signals during DBS surgery. The study analyses 2904 MERs of 26 PD patients who underwent DBS implantation. The low and high order statistical parameters are extracted from the wavelet packet coefficients of MERs and used in the classifications, namely, non-STN vs. STN, pre-STN vs. STN and STN vs. post-STN. RESULTS: Most of the features are significantly different in STN and its subcortical regions, namely, pre-STN and post-STN. The proposed features achieve an average accuracy of 85 % in non-STN vs. STN, 87.2 % in pre-STN vs. STN and 77.7 % in STN vs. post-STN. The accuracy is improved by around 10 % in non-STN vs. STN and STN vs. post-STN when the transition error is 1 mm. COMPARISON WITH EXISTING METHODS: The proposed features are found to be better than the wavelet features. CONCLUSIONS: The proposed approach could be a potential useful adjunct for the real-time rapid intraoperative identification of STN and its anatomical borders.

Could we have missed out the seizure onset: A study based on intracranial EEG.

OBJECTIVE: Intracranial EEG covers only a small fraction of brain volume and it is uncertain if a discharge represents a true seizure onset or results from spread. We therefore assessed if there are differences between characteristics of the ictal onset when we are likely to have a true onset, and characteristics of the discharge in regions of spread. METHODS: Wavelet based statistical features were extracted in 503 onset and 390 spread channels of 58 seizures from 20 patients. These features were used as predictors in models based on machine learning algorithms such as k-nearest neighbour, logistic regression, multilayer perceptron, support vector machine, random and rotation forest. RESULTS: Statistical features (mean, variance, skewness and kurtosis) associated with all wavelet scales were significantly higher in onset than in spread channels. The best classifier, random forest, achieved accuracy of 79.6% and precision of 82%. CONCLUSIONS: The signals associated with onset and spread regions exhibit different characteristics. The proposed features are able to classify the signals with good accuracy. SIGNIFICANCE: Using our classifier on new seizures could help clinicians gain confidence in having recorded the real seizure onset or on the contrary be concerned that the true onset may have been missed.

Detection of Subthalamic Nucleus using Time-Frequency Features of Microelectrode recordings and Random Forest Classifier.

Accurate localization of subthalamic nucleus (STN) is a key prior in deep brain stimulation (DBS) surgery for the patients with advanced Parkinson's disease (PD). Microelectrode recordings (MERs) along with preplanned trajectories are often employed for the STN localization and it remains challenging task. These MER signals are nonstationary and multicomponent in nature. In this study, we propose a system based on time-frequency features of MERs to differentiate the STN and non-STN regions. We assessed the system with 50 MER trajectories from 26 PD patients who have undergone DBS surgery. The signals are pre-processed and subjected to six-level wavelet decomposition. Then, the entropy is computed from the detailed and approximate coefficients. These features are fed to the random forest classifier and the model is evaluated by leave one patient out cross-validation. The results show that entropy associated with detailed wavelet coefficients (D1and D2) are higher in STN where as it is lower in other wavelet scales. All extracted features except entropy from approximate coefficients are found to have significant difference between non-STN and STN (p<; 0.05). The random forest classifier achieves about 83% accuracy and 87% precision in differentiating the STN and non-STN regions.