Search

Deep-learning-based QRS-detection algorithms often require essential post-processing to refine the output prediction-stream for R-peak localisation. The post-processing involves basic signal-processing tasks including the removal of random noise in the model's prediction stream using a basic Salt and Pepper filter, as well as, tasks that use domain-specific thresholds, including a minimum QRS size, and a minimum or maximum R-R distance. These thresholds were found to vary among QRS-detection studies and empirically determined for the target dataset, which may have implications if the target dataset differs such as the drop of performance in unknown test datasets. Moreover, these studies, in general, fail to identify the relative strengths of deep-learning models and the post-processing to weigh them appropriately. This study identifies the domain-specific post-processing, as found in the QRS-detection literature, as three steps based on the required domain knowledge. It was found that the use of minimal domain-specific post-processing is often sufficient for most of the cases and the use of additional domain-specific refinement ensures superior performance, however, it makes the process biased towards the training data and lacks generalisability. As a remedy, a domain-agnostic automated post-processing is introduced where a separate recurrent neural network (RNN)-based model learns required post-processing from the output generated from a QRS-segmenting deep learning model, which is, to the best of our knowledge, the first of its kind. The RNN-based post-processing shows superiority over the domain-specific post-processing for most of the cases (with shallow variants of the QRS-segmenting model and datasets like TWADB) and lags behind for others but with a small margin ( ≤ 2%). The consistency of the RNN-based post-processor is an important characteristic which can be utilised in designing a stable and domain agnostic QRS detector.

Interpretability often seeks domain-specific facts, which is understandable to human, from deep-learning (DL) or other machine-learning (ML) models of black-box nature. This is particularly important to establish transparency in ML model's inner-working and decision-making, so that a certain level of trust is achieved when a model is deployed in a sensitive and mission-critical context, such as health-care. Model-level transparency can be achieved when its components are transparent and are capable of explaining reason of a decision, for a given input, which can be linked to domain-knowledge. This article used convolutional neural network (CNN), with sinc-convolution as its constrained first-layer, to explore if such a model's decision-making can be explained, for a given task, by observing the sinc-convolution's sinc-kernels. These kernels work like band-pass filters, having only two parameters per kernel - lower and upper cutoff frequencies, and optimised through back-propagation. The optimised frequency-bands of sinc-kernels may provide domain-specific insights for a given task. For a given input instance, the effects of sinc-kernels was visualised by means of explanation vector, which may help to identify comparatively significant frequency-bands, that may provide domain-specific interpretation, for the given task. In addition, a CNN model was further optimised by considering the identified subset of prominent sinc frequency-bands as the constrained first-layer, which yielded comparable or better performance, as compared to its all sinc-bands counterpart, as well as, a classical CNN. A minimal CNN structure, achieved through such an optimisation process, may help design task-specific interpretable models. To the best of our knowledge, the idea of sinc-convolution layer's task-specific significant sinc-kernel-based network optimisation is the first of its kind. Additionally, the idea of explanation-vector-based joint time-frequency representation to analyse time-series signals is rare in the literature. The above concept was validated for two tasks, ECG beat-classification (five-class classification task), and R-peak localisation (sample-wise segmentation task).

Electrocardiogram (ECG) signal quality indices (SQIs) are essential for improving diagnostic accuracy and reliability of ECG analysis systems. In various practical applications, the ECG signals are corrupted by different types of noise. These corrupted ECG signals often provide insufficient and incorrect information regarding a patient's health. To solve this problem, signal quality measurements should be made before an ECG signal is used for decision-making. This paper investigates the robustness of existing popular statistical signal quality indices (SSQIs): relative power of QRS complex (SQI p ), skewness (SQI skew ), signal-to-noise ratio (SQI snr ), higher order statistics SQI (SQI hos ) and peakedness of kurtosis (SQI kur ). We analysed the robustness of these SSQIs against different window sizes across diverse datasets. Results showed that the performance of SSQIs considerably fluctuates against varying datasets, whereas the impact of varying window sizes was minimal. This fluctuation occurred due to the use of a static threshold value for classifying noise-free ECG signals from the raw ECG signals. Another drawback of these SSQIs is the bias towards noise-free ECG signals, that limits their usefulness in clinical settings. In summary, the fixed threshold-based SSQIs cannot be used as a robust noise detection system. In order to solve this fixed threshold problem, other techniques can be developed using adaptive thresholds and machine-learning mechanisms.

In this paper we investigate a hybrid model based on the Discrete Gradient method and an evolutionary strategy for determining the weights in a feed forward artificial neural network. Also we discuss different variants for hybrid models using the Discrete Gradient method and an evolutionary strategy for determining the weights in a feed forward artificial neural network. The Discrete Gradient method has the advantage of being able to jump over many local minima and find very deep local minima. However, earlier research has shown that a good starting point for the discrete gradient method can improve the quality of the solution point. Evolutionary algorithms are best suited for global optimisation problems. Nevertheless they are cursed with longer training times and often unsuitable for real world application. For optimisation problems such as weight optimisation for ANNs in real world applications the dimensions are large and time complexity is critical. Hence the idea of a hybrid model can be a suitable option. In this paper we propose different fusion strategies for hybrid models combining the evolutionary strategy with the discrete gradient method to obtain an optimal solution much quicker. Three different fusion strategies are discussed: a linear hybrid model, an iterative hybrid model and a restricted local search hybrid model. Comparative results on a range of standard datasets are provided for different fusion hybrid models.

Entropy measures that assess signals’ complexity have drawn increasing attention recently in biomedical field, as they have shown the ability of capturing unique features that are intrinsic and physiologically meaningful. In this study, we applied entropy analysis to electroencephalogram (EEG) data to examine its performance in epilepsy detection based on short-term EEG, aiming at establishing a short-term analysis protocol with optimal seizure detection performance. Two classification problems were considered, i.e., 1) classifying interictal and ictal EEGs (epileptic group) from normal EEGs; and 2) classifying ictal from interictal EEGs. For each problem, we explored two protocols to analyze the entropy of EEG: i) using a single analytical window with different window lengths, and ii) using an average of multiple windows for each window length. Two entropy methods—fuzzy entropy (FuzzyEn) and distribution entropy (DistEn)–were used that have valid outputs for any given data lengths. We performed feature selection and trained classifiers based on a cross-validation process. The results show that performance of FuzzyEn and DistEn may complement each other and the best performance can be achieved by combining: 1) FuzzyEn of one 5-s window and the averaged DistEn of five 1-s windows for classifying normal from epileptic group (accuracy: 0.93, sensitivity: 0.91, specificity: 0.96); and 2) the averaged FuzzyEn of five 1-s windows and DistEn of one 5-s window for classifying ictal from interictal EEGs (accuracy: 0.91, sensitivity: 0.93, specificity: 0.90). Further studies are warranted to examine whether this proposed short-term analysis procedure can help track the epileptic activities in real time and provide prompt feedback for clinical practices. [ABSTRACT FROM AUTHOR]

This study empirically evaluates the effectiveness of different feature types for the classification of the first language of an author. In particular, it examines the utility of psycholinguistic features, extracted by the Linguistic Inquiry and Word Count ( LIWC) tool, that have not previously been applied to the task of author profiling. As LIWC is a tool that has been developed in the psycholinguistic field rather than the computational linguistics field, it was hypothesized that it would be effective, both as a single type feature set because of its psycholinguistic basis, and in combination with other feature sets, because it should be sufficiently different to add insight rather than redundancy. It was found that LIWC features were competitive with previously used feature types in identifying the first language of an author, and that combined feature sets including LIWC features consistently showed better accuracy rates and average F measures than were achieved by the same feature sets without the LIWC features. As a secondary issue, this study also examined how effectively first language classification scaled up to a larger number of possible languages. It was found that the classification scheme scaled up effectively to the entire 16 language collection from the International Corpus of Learner English, when compared with results achieved on just 5 languages in previous research. [ABSTRACT FROM AUTHOR]