Leave-one-subject-out Cross-validation Research Articles

Calibration trial of an innovative medical device (NEVVA© ) for the evaluation of pain in non-communicating patients in the intensive care unit.

Pain management is an essential and complex issue for non-communicative patients undergoing sedation in the intensive care unit (ICU). The Behavioral Pain Scale (BPS), although not perfect for assessing behavioral pain, is the gold standard based partly on clinical facial expression. NEVVA© , an automatic pain assessment tool based on facial expressions in critically ill patients, is a much-needed innovative medical device. In this prospective pilot study, we recorded the facial expressions of critically ill patients in the medical ICU of Caen University Hospital using the iPhone and Smart Motion Tracking System (SMTS) software with the Facial Action Coding System (FACS) to measure human facial expressions metrically during sedation weaning. Analyses were recorded continuously, and BPS scores were collected hourly over two 8 h periods per day for 3 consecutive days. For this first stage, calibration of the innovative NEVVA© medical device algorithm was obtained by comparison with the reference pain scale (BPS). Thirty participants were enrolled between March and July 2022. To assess the acute severity of illness, the Sequential Organ Failure Assessment (SOFA) and the Simplified Acute Physiology Score (SAPS II) were recorded on ICU admission and were 9 and 47, respectively. All participants had deep sedation, assessed by a Richmond Agitation and Sedation scale (RASS) score of less than or equal to -4 at the time of inclusion. One thousand and six BPS recordings were obtained, and 130 recordings were retained for final calibration: 108 BPS recordings corresponding to the absence of pain and 22 BPS recordings corresponding to the presence of pain. Due to the small size of the dataset, a leave-one-subject-out cross-validation (LOSO-CV) strategy was performed, and the training results obtained the receiver operating characteristic (ROC) curve with an area under the curve (AUC) of 0.792. This model has a sensitivity of 81.8% and a specificity of 72.2%. This pilot study calibrated the NEVVA© medical device and showed the feasibility of continuous facial expression analysis for pain monitoring in ICU patients. The next step will be to correlate this device with the BPS scale.

Adaptive convolutional dictionary learning for denoising seismocardiogram to enhance the classification performance of aortic stenosis

Background:Aortic stenosis (AS) is the most prevalent type of valvular heart disease (VHD), traditionally diagnosed using echocardiogram or phonocardiogram. Seismocardiogram (SCG), an emerging wearable cardiac monitoring modality, is proved to be feasible in non-invasive and cost-effective AS diagnosis. However, SCG waveforms acquired from patients with heart diseases are typically weak, making them more susceptible to noise contamination. While most related researches focus on motion artifacts, sensor noise and quantization noise have been mostly overlooked. These noises pose additional challenges for extracting features from the SCG, especially impeding accurate AS classification. Method:To address this challenge, we present a convolutional dictionary learning-based method. Based on sparse modeling of SCG, the proposed method generates a personalized adaptive-size dictionary from noisy measurements. The dictionary is used for sparse coding of the noisy SCG into a transform domain. Reconstruction from the domain removes the noise while preserving the individual waveform pattern of SCG. Results:Using two self-collected SCG datasets, we established optimal dictionary learning parameters and validated the denoising performance. Subsequently, the proposed method denoised SCG from 50 subjects (25 AS and 25 non-AS). Leave-one-subject-out cross-validation (LOOCV) was applied to 5 machine learning classifiers. Among the classifiers, a bi-layer neural network achieved a moderate accuracy of 90.2%, with an improvement of 13.8% from the denoising. Conclusions:The proposed sparsity-based denoising technique effectively removes stochastic sensor noise and quantization noise from SCG, consequently improving AS classification performance. This approach shows promise for overcoming instrumentation constraints of SCG-based diagnosis.

A novel feature extraction method for the localization of accessory pathways in patients with Wolff-Parkinson-White syndrome

Background and objectiveThe localization of accessory pathways (APs) in patients with Wolff-Parkinson-White (WPW) syndrome using surface electrocardiogram (ECG) is often entirely subjective. This study aimed to develop a novel, more automated, and non-invasive localization method to differentiate right and left APs. Materials and methodsThe participants were 31 patients (aged 8–69 y, mean age: 31.19 ± 14.69 y, 32.3 % women) with manifest WPW syndrome who were treated successfully in the first ablation session. The novel localization of APs was based on feature extraction using McSharry’s model, and SFS-LOV, a combination of three methods (sequential forward selection (SFS), leave-one-subject-out cross-validation (LOSO), and majority voting). The k-nearest neighbor (KNN) and support vector machine (SVM) classifier were separately used in the SFS-LOV method. In this method, the features were extracted from three segments of the ECG signals. Each segment was started from the onset of the P-wave and terminated with the offset of the T-wave. The majority voting was applied to the results of the classifier on the segments of each participant. ResultsThe proposed method differentiated the right from the left APs with an accuracy of 87% (sensitivity: 80%, specificity: 94%). This was achieved by feeding four selected parameters of McSharry’s model into the KNN classifier. These selected parameters were estimated from the lead V4 of the ECG signals. ConclusionThe parameters of McSharry’s model can semi-automatically localize the right and left APs in patients with manifest WPW syndrome.

Recognizing hand use and hand role at home after stroke from egocentric video.

Hand function is a central determinant of independence after stroke. Measuring hand use in the home environment is necessary to evaluate the impact of new interventions, and calls for novel wearable technologies. Egocentric video can capture hand-object interactions in context, as well as show how more-affected hands are used during bilateral tasks (for stabilization or manipulation). Automated methods are required to extract this information. The objective of this study was to use artificial intelligence-based computer vision to classify hand use and hand role from egocentric videos recorded at home after stroke. Twenty-one stroke survivors participated in the study. A random forest classifier, a SlowFast neural network, and the Hand Object Detector neural network were applied to identify hand use and hand role at home. Leave-One-Subject-Out-Cross-Validation (LOSOCV) was used to evaluate the performance of the three models. Between-group differences of the models were calculated based on the Mathews correlation coefficient (MCC). For hand use detection, the Hand Object Detector had significantly higher performance than the other models. The macro average MCCs using this model in the LOSOCV were 0.50 ± 0.23 for the more-affected hands and 0.58 ± 0.18 for the less-affected hands. Hand role classification had macro average MCCs in the LOSOCV that were close to zero for all models. Using egocentric video to capture the hand use of stroke survivors at home is technically feasible. Pose estimation to track finger movements may be beneficial to classifying hand roles in the future.

Limitations of Pacemaker Spike Detection in Capacitive ECGs via Deep Learning

Abstract Pacemaker spike detection is an important step in monitoring paced patients. Capacitive ECG facilitates unobtrusive monitoring of subjects during daily routines such as driving. Robust algorithms are required to deal with low signal quality and artifacts, e.g. by employing fusion of multiple signal channels. Due to the low signal-to-noise ratio of the measurement, there are limitations to detection accuracy compared to conventional ECG monitors. Especially low voltage stimulations such as bipolar pacemaker spikes are hard to detect. We present a convolutional network approach to improve on recent signal processing algorithms.We show a realistic evaluation of its performance using leave-one-subject-out cross validation (LOOCV), its dependence on the size of the receptive field, and an estimation of an upper performance bound.

EEG-based clinical decision support system for Alzheimer's disorders diagnosis using EMD and deep learning techniques.

Despite the existence of numerous clinical techniques for identifying neurological brain disorders in their early stages, Electroencephalogram (EEG) data shows great promise as a means of detecting Alzheimer's disease (AD) at an early stage. The main goal of this research is to create a reliable and accurate clinical decision support system leveraging EEG signal processing to detect AD in its initial phases. The research utilized a dataset consisting of 35 neurotypical individuals, 31 patients with mild AD, and 22 patients with moderate AD. Data were collected while participants were at rest. To extract features from the EEG signals, a band-pass filter was applied to the dataset and the Empirical Mode Decomposition (EMD) technique was employed to decompose the filtered signals. The EMD technique was then leveraged to generate feature vectors by combining multiple signal features, thereby enhancing diagnostic performance. Various artificial intelligence approaches were also explored and compared to identify features of the extracted EEG signals distinguishing mild AD, moderate AD, and neurotypical cases. The performance of the classifiers was evaluated using k-fold cross-validation and leave-one-subject-out (LOSO) cross-validation methods. The results of this study provided valuable insights into potential avenues for the early diagnosis of AD. The performance of the various offered methodologies has been compared and evaluated by computing the overall diagnosis precision, recall, and accuracy. The proposed methodologies achieved a maximum classification accuracy of 99.9 and 94.8% for k-fold and LOSO cross-validation techniques, respectively. The study aims to assess and compare different proposed methodologies and determine the most effective combination approach for the early detection of AD. Our research findings strongly suggest that the proposed diagnostic support technique is a highly promising supplementary tool for discovering prospective diagnostic biomarkers that can greatly aid in the early clinical diagnosis of AD.

EEG rhythm based emotion recognition using multivariate decomposition and ensemble machine learning classifier

Recently, electroencephalogram (EEG) signals have shown great potential to recognize human emotions. The goal of effective computing is to assist computers in understanding various types of emotions via human–computer interaction (HCI). Multichannel EEG signals are used to measure the electrical activity of the brain in space and time. Automated emotion recognition using multichannel EEG signals is an interesting area of cognitive neuroscience and affective computing research. This research proposes EEG multichannel rhythmic features and ensemble machine learning (EML) classifiers with leave-one-subject-out cross-validation (LOSOCV) for automatic emotion classification from multichannel EEG recordings. Multivariate fast iterative filtering (MvFIF) is used to assess the EEG rhythm sequences. EEG rhythms delta(δ), theta(θ), alpha(α), beta(β), and gamma(γ) are separated based on the mean frequency of the EEG rhythm sequence. Three Hjorth parameters and nine entropy features were extracted from multichannel EEG rhythms. Extracted features are selected using the minimum redundancy maximum relevance (mRMR) approach. The experimental design was performed on two emotional datasets (GAMEEMO and DREAMER). The validation showed that gamma rhythm multichannel features with EML-based subspace K-nearest neighbor (SS KNN) were as high as 93.5%–99.8%, achieving high classification accuracy. The comparisons of δ, θ, α, β, and γ rhythms with EML, support vector machine (SVM), and artificial neural network (ANN) were performed. we also analyzed multi-class emotions (HVHA, HVLA, LVHA, LVLA) with an ensemble-based bagging tree on gamma rhythm. It provides a novel solution for multichannel rhythm-specific features in EEG data analysis.

Epilepsy detection in 121 patient populations using hypercube pattern from EEG signals

BackgroundEpilepsy is one of the most commonly seen neurologic disorders worldwide and has generally caused seizures. Electroencephalography (EEG) is widely used in seizure diagnosis. To detect epilepsy automatically, various machine learning (ML) models have been introduced in the literature, but the used EEG signal datasets for epilepsy detection are relatively small. Our main objective is to present a large EEG signal dataset and investigate the detection ability of a new hypercube pattern-based framework using the EEG signals. Material and methodThis study collected a large EEG signal dataset (10,356 EEG signals) from 121 participants. We proposed a new information fusion-based feature engineering framework to get high classification performance from this dataset. The dataset consists of 35 channels, and our proposed feature engineering model extracts features from each channel. A new hypercube-based feature extractor has been proposed to generate two feature vectors in the feature extraction phase. Various statistical parameters of the signals have been used to create a feature vector. Multilevel discrete wavelet transform (MDWT) has been applied to develop a multileveled feature extraction function, and seven feature vectors have been extracted. In this work, we have extracted 245 (=35 × 7) feature vectors, and the most valuable features from these vectors have been selected using the neighborhood component analysis (NCA) selector. Finally, these selected features were fed to the k nearest neighbors (kNN) classifier with the leave one subject out (LOSO) cross-validation (CV) strategy. These results have been voted/fused to obtain the highest classification performance. ResultsIn this work, we have attained 87.78% classification accuracy using voting these vectors and 79.07% with LOSO CV with the EEG signals. ConclusionsThe proposed fusion-based feature engineering model achieved satisfactory classification performance using the largest EEG signal datasets for epilepsy detection.

Unveiling the Relationships Between Seismocardiogram Signals, Physical Activity Types and Metabolic Equivalent of Task Scores.

The diagnosis of metabolic syndrome and cardiovascular disorders can highly benefit from physical activity and energy expenditure assessment. In this study, we investigated the relationship between metabolic equivalent of task (MET) scores and seismocardiogram (SCG)-derived parameters. We worked with the PAMAP2 dataset and focused on the 3-axial chest acceleration data. We first segmented the 3-axial SCG signals into respiration (0-1Hz), cardiac vibrations (1-20Hz) and heart sounds (20-40Hz) components. Additionally, we investigated their combinations: 0-20Hz, 1-40Hz and 0-40Hz. We then windowed each signal, and extracted time and frequency domain features from each window. Using the MET scores and activity types, we trained linear regression and random forest classification models first using 80-20% split, then with leave-one-subject-out cross-validation (LOSO-CV). Additionally, we investigated the significance of each feature and axis. For the 80-20% task, the best performing frequency bands were 0-1Hz, 0-20Hz, and 0-40Hz, which yielded a (MET mean-squared-error, classification accuracy) pair of (0.354, 0.952), (0.367, 0.904), and (0.377, 0.914), respectively. When LOSO-CV was applied, we obtained (1.059, 0.865), (0.681, 0.868), and (0.804, 0.875) for each band, respectively. Additionally, our results revealed that the lateral axis provides the most critical information about cardiorespiratory effect of performed activities. Different SCG components can provide unique and substantial contributions to activity and energy expenditure assessment. This framework can be leveraged in the design of wearable systems for monitoring the activity and energy expenditure levels, and understanding their relationship with underlying cardiorespiratory parameters.

An End-to-End Deep Learning Model for EEG-Based Major Depressive Disorder Classification

Major depressive disorder (MDD) is a prevalent mental illness associated with abnormalities in structural and functional brain connectivity, and it has become a global public health problem. Early diagnosis is important and challenging for the treatment of MDD. Previous studies proposed the classification methods for MDD based on brain connectivity features through functional connectivity (FC) or effective connectivity (EC) measures. However, it requires prior knowledge and experience to manually select a algorithm to calculate the brain connectivity features. Given that the representation learning capabilities of deep learning (DL) models and the ability to capture correlations between data of self-attention mechanism, we proposed an end-to-end integrated DL model for classifying MDD patients and healthy controls (HCs) based on resting-state electroencephalography (EEG) data. This model first automatically learned the potential connectivity relationships among EEG channels through a multi-head self-attention mechanism, and then extracted higher-level features through a parallel two-branch convolution neural network (CNN) module, and finally completed the classification through a fully connected layer. A public resting-state EEG dataset was utilized to evaluate the validity of the proposed model. The experimental results indicated that the proposed model achieved 91.06% average accuracy that was better than those of comparison methods using the leave-one-subject-out cross-validation (LOSOCV) method. This study may provide a novel approach for brain connectivity modeling of MDD detection.

An Automatic Premature Ventricular Contraction Recognition System Based on Imbalanced Dataset and Pre-Trained Residual Network Using Transfer Learning on ECG Signal

The development of automatic monitoring and diagnosis systems for cardiac patients over the internet has been facilitated by recent advancements in wearable sensor devices from electrocardiographs (ECGs), which need the use of patient-specific approaches. Premature ventricular contraction (PVC) is a common chronic cardiovascular disease that can cause conditions that are potentially fatal. Therefore, for the diagnosis of likely heart failure, precise PVC detection from ECGs is crucial. In the clinical settings, cardiologists typically employ long-term ECGs as a tool to identify PVCs, where a cardiologist must put in a lot of time and effort to appropriately assess the long-term ECGs which is time consuming and cumbersome. By addressing these issues, we have investigated a deep learning method with a pre-trained deep residual network, ResNet-18, to identify PVCs automatically using transfer learning mechanism. Herein, features are extracted by the inner layers of the network automatically compared to hand-crafted feature extraction methods. Transfer learning mechanism handles the difficulties of required large volume of training data for a deep model. The pre-trained model is evaluated on the Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia and Institute of Cardiological Technics (INCART) datasets. First, we used the Pan-Tompkins algorithm to segment 44,103 normal and 6423 PVC beats, as well as 106,239 normal and 9987 PVC beats from the MIT-BIH Arrhythmia and IN-CART datasets, respectively. The pre-trained model employed the segmented beats as input after being converted into 2D (two-dimensional) images. The method is optimized with the using of weighted random samples, on-the-fly augmentation, Adam optimizer, and call back feature. The results from the proposed method demonstrate the satisfactory findings without the using of any complex pre-processing and feature extraction technique as well as design complexity of model. Using LOSOCV (leave one subject out cross-validation), the received accuracies on MIT-BIH and INCART are 99.93% and 99.77%, respectively, suppressing the state-of-the-art methods for PVC recognition on unseen data. This demonstrates the efficacy and generalizability of the proposed method on the imbalanced datasets. Due to the absence of device-specific (patient-specific) information at the evaluating stage on the target datasets in this study, the method might be used as a general approach to handle the situations in which ECG signals are obtained from different patients utilizing a variety of smart sensor devices.

Automated detection of construction work at heights and deployment of safety hooks using IMU with a barometer

An automated system that identifies work at height and the fastening state of safety hooks using wearable sensors was developed to prevent falls from height (FFH). This system estimates the altitudes of workers based on the atmospheric pressure measured by a barometer and acceleration and gyroscopic signals from an inertial measurement unit (IMU). The fastening state of the safety hooks of workers at height is determined with the data collected by the IMU sensor and machine learning algorithms. Although researchers have tried to detect unsafe work conditions and unsafe behaviors at height, the complicated tasks and dynamic work conditions have discouraged them from establishing precise methodologies for effective and timely detection. To validate the system of this study, on-site field experiments were conducted to collect data from 20 construction workers. The performance of the developed model was assessed with leave-one-subject-out cross-validation (LOSOCV) to accommodate a wide range of new workers and their working conditions. According to the results, the work-at-height identification system is 96% accurate, while the safety hook attachment detection system is 86% accurate. The findings of this study fill knowledge gaps by providing ways of identifying workers working at height and detecting the fastening state of safety hooks in a non-invasive and objective manner. The results are expected to improve safety management at construction sites by minimizing the FFH risk for workers working at height.

Improving automated diagnosis of epilepsy from EEGs beyond IEDs

Objective. Clinical diagnosis of epilepsy relies partially on identifying interictal epileptiform discharges (IEDs) in scalp electroencephalograms (EEGs). This process is expert-biased, tedious, and can delay the diagnosis procedure. Beyond automatically detecting IEDs, there are far fewer studies on automated methods to differentiate epileptic EEGs (potentially without IEDs) from normal EEGs. In addition, the diagnosis of epilepsy based on a single EEG tends to be low. Consequently, there is a strong need for automated systems for EEG interpretation. Traditionally, epilepsy diagnosis relies heavily on IEDs. However, since not all epileptic EEGs exhibit IEDs, it is essential to explore IED-independent EEG measures for epilepsy diagnosis. The main objective is to develop an automated system for detecting epileptic EEGs, both with or without IEDs. In order to detect epileptic EEGs without IEDs, it is crucial to include EEG features in the algorithm that are not directly related to IEDs. Approach. In this study, we explore the background characteristics of interictal EEG for automated and more reliable diagnosis of epilepsy. Specifically, we investigate features based on univariate temporal measures (UTMs), spectral, wavelet, Stockwell, connectivity, and graph metrics of EEGs, besides patient-related information (age and vigilance state). The evaluation is performed on a sizeable cohort of routine scalp EEGs (685 epileptic EEGs and 1229 normal EEGs) from five centers across Singapore, USA, and India. Main results. In comparison with the current literature, we obtained an improved Leave-One-Subject-Out (LOSO) cross-validation (CV) area under the curve (AUC) of 0.871 (Balanced Accuracy (BAC) of 80.9%) with a combination of three features (IED rate, and Daubechies and Morlet wavelets) for the classification of EEGs with IEDs vs. normal EEGs. The IED-independent feature UTM achieved a LOSO CV AUC of 0.809 (BAC of 74.4%). The inclusion of IED-independent features also helps to improve the EEG-level classification of epileptic EEGs with and without IEDs vs. normal EEGs, achieving an AUC of 0.822 (BAC of 77.6%) compared to 0.688 (BAC of 59.6%) for classification only based on the IED rate. Specifically, the addition of IED-independent features improved the BAC by 21% in detecting epileptic EEGs that do not contain IEDs. Significance. These results pave the way towards automated detection of epilepsy. We are one of the first to analyze epileptic EEGs without IEDs, thereby opening up an underexplored option in epilepsy diagnosis.

Subject-Independent Classification of P300 Event-Related Potentials Using a Small Number of Training Subjects

The intersubject variability present in electroencephalographic (EEG) signals can affect the performance of the brain–computer interface (BCI) systems. Despite the significant progress in the field, the variability in neural data remains one of the most critical challenges in constructing accurate predictive models of human intention. As a result, the majority of the previous studies have focused either on devising subject-specific signal processing and machine learning algorithms, used some data from a target user to update and calibrate a pretrained classifier, or have used data collected from a relatively large number of training subjects to construct generic classifiers for new subjects. In this work, we investigate the feasibility of using a relatively small number of training subjects to achieve subject-independent classification of event-related potentials (ERPs) in P300-based BCIs. To this end, we employ convolutional neural networks (CNNs) and propose a leave-one-subject-out cross-validation (LOSO-CV) for model selection; that is to say, for tuning CNN hyperparameters including number of layers, filters, kernel size, and epoch. The utility of the proposed model selection is warranted because LOSO-CV simulates the effect of subject-independent classification within the training data. The entire process of training (including model selection) is validated by applying another LOSO-CV external to the training process. Our empirical results obtained on four publicly available datasets confirm the capability of LOSO-CV model selection with CNN to capture intrinsic ERP features from a small group of subjects to classify observations collected from unseen subjects.

A Deep Learning Framework for Automatic Sleep Apnea Classification Based on Empirical Mode Decomposition Derived from Single-Lead Electrocardiogram.

Background: Although polysomnography (PSG) is a gold standard tool for diagnosing sleep apnea (SA), it can reduce the patient’s sleep quality by the placement of several disturbing sensors and can only be interpreted by a highly trained sleep technician or scientist. In recent years, electrocardiogram (ECG)-derived respiration (EDR) and heart rate variability (HRV) have been used to automatically diagnose SA and reduce the drawbacks of PSG. Up to now, most of the proposed approaches focus on machine-learning (ML) algorithms and feature engineering, which require prior expert knowledge and experience. The present study proposes an SA detection algorithm to differentiate a normal and apnea event using a deep-learning (DL) framework based on 1D and 2D deep CNN with empirical mode decomposition (EMD) of a preprocessed ECG signal. The EMD is ideally suited to extract essential components which are characteristic of the underlying biological or physiological processes. In addition, the simple and compact architecture of 1D deep CNN, which only performs 1D convolutions, and pretrained 2D deep CNNs, are suitable for real-time and low-cost hardware implementation. Method: This study was validated using 7 h to nearly 10 h overnight ECG recordings from 33 subjects with an average apnea-hypopnea index (AHI) of 30.23/h originated from PhysioNet Apnea-ECG database (PAED). In preprocessing, the raw ECG signal was normalized and filtered using the FIR band pass filter. The preprocessed ECG signal was then decomposed using the empirical mode decomposition (EMD) technique to generate several features. Several important generated features were selected using neighborhood component analysis (NCA). Finally, deep learning algorithm based on 1D and 2D deep CNN were used to perform the classification of normal and apnea event. The synthetic minority oversampling technique (SMOTE) was also applied to evaluate the influence of the imbalanced data problem. Results: The segment-level classification performance had 93.8% accuracy with 94.9% sensitivity and 92.7% specificity based on 5-fold cross-validation (5fold-CV), meanwhile, the subject-level classification performance had 83.5% accuracy with 75.9% sensitivity and 88.7% specificity based on leave-one-subject-out cross-validation (LOSO-CV). Conclusion: A novel and robust SA detection algorithm based on the ECG decomposed signal using EMD and deep CNN was successfully developed in this study.

EEG-based cross-subject emotion recognition using Fourier-Bessel series expansion based empirical wavelet transform and NCA feature selection method

Automated emotion recognition using brain electroencephalogram (EEG) signals is predominantly used for the accurate assessment of human actions as compared to facial expression or speech signals. Various signal processing methods have been used for extracting representative features from EEG signals for emotion recognition. However, the EEG signals are non-stationary and vary across the subjects as well as in different sessions of the same subject; hence it exhibits poor generalizability and low classification accuracy for an emotion classification of cross subjects. In this paper, EEG signals-based automated cross-subject emotion recognition framework is proposed using the Fourier-Bessel series expansion-based empirical wavelet transform (FBSE-EWT) method. This method is used to decompose the EEG signals from each channel into four sub-band signals. Manually ten channels are selected from the frontal lobe, from which entropy and energy features are extracted from each sub-band signal. The subject variability is reduced using an average moving filter method on each channel to obtain the smoothened feature vector of size 80. The three feature selection techniques, such as neighborhood component analysis (NCA), relief-F, and mRMR, are used to obtain an optimal feature vector. The machine learning models, such as artificial neural network (ANN), k-nearest neighborhood (k-NN) with two (fine and weighted) functions, and ensemble bagged tree classifiers are trained by the obtained feature vectors. The experiments are performed on two publicly accessible databases, named SJTU emotion EEG dataset (SEED) and dataset for emotion analysis using physiological signals (DEAP). The training and testing of the models have been performed using 10-fold cross-validation and leave-one-subject-out-cross-validation (LOSOCV). The proposed framework based on FBSE-EWT and NCA feature selection approach shows superior results for classifying human emotions compared to other state-of-art emotion classification models.

Are Gyroscopes an Added Value in Leave-One-Subject-Out Activity Recognition with IMUs?

Inertial sensors have played a key role in the development of Human Activity Recognition (HAR) systems. Adding gyroscopes in HAR systems leads to increased battery and processing resources. Therefore, it is important to explore their added value compared with using accelerometers only. This study evaluates the added value of gyroscopes in activity recognition. Two public available datasets recorded by accelerometers and gyroscopes were studied. These datasets focus on multiple types of activities: UCI HAR dataset includes walking, walking upstairs, walking downstairs, sitting, standing, laying and WISDM dataset includes 18 hand-oriented and non-hand-oriented activities. Several machine learning models were applied to both datasets for activity recognition. Leave-one-subject-out cross-validation (LOSO) was applied to evaluate the models, where the training set and test set were from different subjects. For UCI HAR dataset, the multilayer perceptron (MLP) model obtained the highest f1-scores. Adding a gyroscope on the waist significantly improved the f1-scores of sitting and laying (both ). For WISDM dataset, the support vector machines (SVM) model obtained the highest f1-scores. The gyroscope on the wrist improved hand-oriented activities while the gyroscope in the pockets improved non-hand-oriented activities (all . The results showed the improvement for recognition performance by adding gyroscopes. However, the improvement was dependent on the type of activity and the mounting place of the gyroscope. Clinical relevance- Gyroscopes are common sensors for activity recognition in wearable healthcare systems. This study proves the added value by adding gyroscopes on different mounting places for recognition performance.

Affect recognition from scalp-EEG using channel-wise encoder networks coupled with geometric deep learning and multi-channel feature fusion

The expression of human emotions is a complex process that often manifests through physiological and psychological traits and results in spatio-temporal brain activity. The brain activity can be captured with an electroencephalogram (EEG) and can be used for emotion recognition. In this paper, we present a novel approach to EEG-based emotion recognition (in terms of arousal, valence, and dominance) using unprocessed EEG signals. Input EEG samples are passed through channel-specific encoders consisting of SincNet based convolution blocks (filters are fine-tuned for the emotion recognition during learning) to learn high-level features related to the objectives. The resultant feature embeddings are then passed through a set of graph convolution networks to model the spatial propagation of brain activity under the assumption that the brain activity captured through an electrode is impacted by the brain activity captured by neighbouring electrodes. The channels are represented as nodes in a graph following the relative positioning of the electrodes during dataset acquisition. Multi-head attention is applied together with the graph convolutions to jointly attend to features from different representation sub-spaces, which leads to improved learning. The resultant features are then passed through a deep neural network-based multi-task classifier to identify the dimensional emotional states (low/high). Our proposed model achieves an accuracy of 88.24%, 88.80% and 88.22% for arousal, valence and dominance respectively using a 10-fold cross-validation; and 63.71%, 64.98% and 61.81% with Leave-One-Subject-Out cross-validation (LOSO) on the Dreamer dataset, and 69.72%, 69.43% and 70.72% for a LOSO evaluation on the DEAP dataset, surpassing state-of-the-art methods.

An accurate valvular heart disorders detection model based on a new dual symmetric tree pattern using stethoscope sounds

Background and purposeValvular heart disease (VHD) is an important cause of morbidity and mortality. Echocardiography is the reference standard for VHD diagnosis but is not universally accessible. Manual cardiac auscultation is inadequate for screening VHD. Many machine learning models using heart sounds acquired with an electronic stethoscope may improve the accuracy of VHD diagnosis. We aimed to develop an accurate sound classification model for VHD diagnosis. Materials and methodsA new large stethoscope sound dataset containing 10,366 heart sounds divided into ten categories (nine VHD and one healthy) were prospectively collected. We developed a handcrafted learning model that comprised multilevel feature extraction based on a dual symmetric tree pattern (DSTP) and multilevel discrete wavelet transform (DWT), feature selection, and classification. The multilevel DWT was used to create subbands to extract features at both high and low levels. Then, iterative neighborhood component analysis was used to select the most discriminative 512 features from among the extracted features in the generated feature vector. In the classification phase, a support vector machine (SVM) was used with 10-fold cross-validation (CV) and leave-one-subject-out (LOSO) CV. ResultsOur proposed DSTP-based model attained 99.58% and 99.84% classification accuracies using SVM classifier with 10-fold CV and LOSO CV, respectively. ConclusionsThe presented DSTP-based classification model attained excellent multiclass classification performance on a large prospective heart sound dataset at a low computational cost.

Hand gesture classification using time–frequency images and transfer learning based on CNN

Hand gesture-based systems are one of the most effective technological advances and continue to develop with improvements in the field of human–computer interaction. Surface electromyography (sEMG) is utilized as a popular input data for gesture classification with elevated accuracy and advanced control capability. This paper presents a comparative hand gesture classification approach using time–frequency (TF) images of the spontaneous sEMG signals and the transfer learning method. 4-channel sEMG signals are collected from 30 subjects performing 7 specific hand gestures. After the required pre-processing, segmentation, and windowing steps, three TF analysis methods, namely Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Hilbert-Huang Transform (HHT), are applied to EMG signals to obtain TF images. Spectrograms from STFT, scalograms from CWT, and Hilbert-Huang spectra (HHS) from HHT obtained from multi-channel sEMG data are separately fused. TF images are then utilized to extract distinct features using seven state-of-the-art, pre-trained Convolutional Neural Network (CNN) architectures and classify seven hand gestures. Two different robust cross-validation strategies are conducted to evaluate the proposed method; stratified k-fold cross-validation (SKCV) and leave-one-subject-out cross-validation (LOOCV). We also investigate the effect of window size and the combination of Intrinsic Mode Functions (IMFs) on classification performance. The results demonstrated that the HHT utilizing IMFs obtained by Empirical Mode Decomposition (EMD) provided improved TF resolution and better results than STFT and CWT in the classification of sEMG signals. Finally, the best average accuracies (93.75% for SKCV) and (94.41% for LOOCV) are obtained by the HHT method with the pre-trained ResNet-50 model.