Machine learning model with output correction: Towards reliable bradycardia detection in neonates

Bradycardia is a commonly occurring condition in premature infants, often causing serious consequences and cardiovascular complications. Reliable and accurate detection of bradycardia events is pivotal for timely intervention and effective treatment. Excessive false alarms pose a critical problem in bradycardia event detection, eroding trust in machine learning (ML)-based clinical decision support tools designed for such detection. This could result in disregarding the algorithm's accurate recommendations and disrupting workflows, potentially compromising the quality of patient care. This article introduces an ML-based approach incorporating an output correction element, designed to minimise false alarms. The approach has been applied to bradycardia detection in preterm infants. We applied five ML-based autoencoder techniques, using recurrent neural network (RNN), long-short-term memory (LSTM), gated recurrent unit (GRU), 1D convolutional neural network (1D CNN), and a combination of 1D CNN and LSTM. The analysis is performed on ∼440 hours of real-time preterm infant data. The proposed approach achieved 0.978, 0.73, 0.992, 0.671 and 0.007 in AUC-ROC, AUC-PRC, recall, F1 score, and false positive rate (FPR) respectively and a false alarms reduction of 36% when compared with methods without the correction approach. This study underscores the imperative of cultivating solutions that alleviate alarm fatigue and encourage active engagement among healthcare professionals.

Modelling methicillin-resistant Staphylococcus aureus decolonization: interactions between body sites and the impact of site-specific clearance

Methicillin-resistant Staphylococcus aureus (MRSA) can colonize multiple body sites, and carriage is a risk factor for infection. Successful decolonization protocols reduce disease incidence; however, multiple protocols exist, comprising diverse therapies targeting multiple body sites, and the optimal protocol is unclear. Standard methods cannot infer the impact of site-specific components on successful decolonization. Here, we formulate a Bayesian coupled hidden Markov model, which estimates interactions between body sites, quantifies the contribution of each therapy to successful decolonization, and enables predictions of the efficacy of therapy combinations. We applied the model to longitudinal data from a randomized controlled trial (RCT) of an MRSA decolonization protocol consisting of chlorhexidine body and mouthwash and nasal mupirocin. Our findings (i) confirmed nares as a central hub for MRSA colonization and nasal mupirocin as the most crucial therapy and (ii) demonstrated all components contributed significantly to the efficacy of the protocol and the protocol reduced self-inoculation. Finally, we assessed the impact of hypothetical therapy improvements in silico and found that enhancing MRSA clearance at the skin would yield the largest gains. This study demonstrates the use of advanced modelling to go beyond what is typically achieved by RCTs, enabling evidence-based decision-making to streamline clinical protocols.

BI - Directional long short-term memory for automatic detection of sleep apnea events based on single channel EEG signal

Sleep apnea syndrome (SAS) is a sleeping disorder in which breathing stops regularly. Even though its prevalence is high, many cases are not reported due to the high cost of inspection and the limits of monitoring devices. To address this, based on the bidirectional long and short-term memory network (BI-LSTM), we designed a single-channel electroencephalography (EEG) sleep monitoring model that can be used in portable SAS monitoring devices. Model training and evaluation of EEG signals obtained by polysomnography were performed on the event segments of 42 subjects. Adam and 10-fold cross-validation were employed to optimize parameters and evaluate network performance. The results showed that BI-LSTM has a precision of 84.21% and accuracy of 92.73%.

Early bradycardia detection and therapeutic interventions in preterm infant monitoring

In very preterm infants, cardio-respiratory events and associated hypoxemia occurring during early postnatal life have been associated with risks of retinopathy, growth alteration and neurodevelopment impairment. These events are commonly detected by continuous cardio-respiratory monitoring in neonatal intensive care units (NICU), through the associated bradycardia. NICU nurse interventions are mainly triggered by these alarms. In this work, we acquired data from 52 preterm infants during NICU monitoring, in order to propose an early bradycardia detector which is based on a decentralized fusion of three detectors. The main objective is to improve automatic detection under real-life conditions without altering performance with respect to that of a monitor commonly used in NICU. We used heart rate lower than 80 bpm during at least 10 sec to define bradycardia. With this definition we observed a high rate of false alarms (64%) in real-life and that 29% of the relevant alarms were not followed by manual interventions. Concerning the proposed detection method, when compared to current monitors, it provided a significant decrease of the detection delay of 2.9 seconds, without alteration of the sensitivity (97.6% vs 95.2%) and false alarm rate (63.7% vs 64.1%). We expect that such an early detection will improve the response of the newborn to the intervention and allow for the development of new automatic therapeutic strategies which could complement manual intervention and decrease the sepsis risk.

Automatic Detection of Obstructive Sleep Apnea Events Using a Deep CNN-LSTM Model

Obstructive sleep apnea (OSA) is a common sleep-related respiratory disorder. Around the world, more and more people are suffering from OSA. Because of the limitation of monitor equipment, many people with OSA remain undetected. Therefore, we propose a sleep-monitoring model based on single-channel electrocardiogram using a convolutional neural network (CNN), which can be used in portable OSA monitor devices. To learn different scale features, the first convolution layer comprises three types of filters. The long short-term memory (LSTM) is used to learn the long-term dependencies such as the OSA transition rules. The softmax function is connected to the final fully connected layer to obtain the final decision. To detect a complete OSA event, the raw ECG signals are segmented by a 10 s overlapping sliding window. The proposed model is trained with the segmented raw signals and is subsequently tested to evaluate its event detection performance. According to experiment analysis, the proposed model exhibits Cohen's kappa coefficient of 0.92, a sensitivity of 96.1%, a specificity of 96.2%, and an accuracy of 96.1% with respect to the Apnea-ECG dataset. The proposed model is significantly higher than the results from the baseline method. The results prove that our approach could be a useful tool for detecting OSA on the basis of a single-lead ECG.

Detection of Apnea Bradycardia from ECG Signals of Preterm Infants Using Layered Hidden Markov Model

Apnea-bradycardia (AB) is a common complication in prematurely born infants, which is associated with reduced survival and neurodevelopmental outcomes. Thus, early detection or predication of AB episodes is critical for initiating preventive interventions. To develop automatic real-time operating systems for early detection of AB, recent advances in signal processing can be employed. Hidden Markov Models (HMM) are probabilistic models with the ability of learning different dynamics of the real time-series such as clinical recordings. In this study, a hierarchy of HMMs named as layered HMM was presented to detect AB episodes from pre-processed single-channel Electrocardiography (ECG). For training the hierarchical structure, RR interval, and width of QRS complex were extracted from ECG as observations. The recordings of 32 premature infants with median 31.2 (29.7, 31.9) weeks of gestation were used for this study. The performance of the proposed layered HMM was evaluated in detecting AB. The best average accuracy of 97.14 ± 0.31% with detection delay of - 5.05 ± 0.41 s was achieved. The results show that layered structure can improve the performance of the detection system in early detecting of AB episodes. Such system can be incorporated for more robust long-term monitoring of preterm infants.

A primer on coupled state-switching models for multiple interacting time series

State-switching models such as hidden Markov models or Markov-switching regression models are routinely applied to analyse sequences of observations that are driven by underlying non-observable states. Coupled state-switching models extend these approaches to address the case of multiple observation sequences whose underlying state variables interact. In this article, we provide an overview of the modelling techniques related to coupling in state-switching models, thereby forming a rich and flexible statistical framework particularly useful for modelling correlated time series. Simulation experiments demonstrate the relevance of being able to account for an asynchronous evolution as well as interactions between the underlying latent processes. The models are further illustrated using two case studies related to (a) interactions between a dolphin mother and her calf as inferred from movement data and (b) electronic health record data collected on 696 patients within an intensive care unit.

Recovering Truncated Streaming Event Log Using Coupled Hidden Markov Model

Process discovery is a technique for obtaining process model based on traces recorded in the event log. Nowadays, information systems produce streaming event logs to record their huge processes. The truncated streaming event log is a big issue in process discovery because it inflicts incomplete traces that make process discovery depict wrong processes in a process model. Earlier research suggested several methods for recovering the truncated streaming event log and none of them utilized Coupled Hidden Markov Model. This research proposes a method that combines Coupled Hidden Markov Model with Double States and the Modification of Viterbi–Backward method for recovering the truncated streaming event log. The first layer of states contains the transition probability of activities. The second layer of states uses patterns for detecting traces which have a low appearance in the event log. The experiment results showed that the proposed method recovered appropriately the truncated streaming event log. These results also have proven that the accuracies of recovered traces obtained by the proposed method are higher than those obtained by the Hidden Markov Model and the Coupled Hidden Markov Model.

Variational Inference for Coupled Hidden Markov Models Applied to the Joint Detection of Copy Number Variations

Hidden Markov models provide a natural statistical framework for the detection of the copy number variations (CNV) in genomics. In this context, we define a hidden Markov process that underlies all individuals jointly in order to detect and to classify genomics regions in different states (typically, deletion, normal or amplification). Structural variations from different individuals may be dependent. It is the case in agronomy where varietal selection program exists and species share a common phylogenetic past. We propose to take into account these dependencies inthe HMM model. When dealing with a large number of series, maximum likelihood inference (performed classically using the EM algorithm) becomes intractable. We thus propose an approximate inference algorithm based on a variational approach (VEM), implemented in the CHMM R package. A simulation study is performed to assess the performance of the proposed method and an application to the detection of structural variations in plant genomes is presented.

Automatic Detection of Obstructive Sleep Apnea Using Wavelet Transform and Entropy-Based Features From Single-Lead ECG Signal

Obstructive sleep apnea (OSA) is a prevalent sleep disorder and highly affects the quality of human life. Currently, gold standard for OSA detection is polysomnogram. Since this method is time consuming and cost inefficient, practical systems focus on the usage of electrocardiogram (ECG) signals for OSA detection. In this paper, a novel automatic OSA detection method using a single-lead ECG signal has been proposed. A nonlinear feature extraction using wavelet transform (WT) coefficients obtained by an ECG signal decomposition is employed. In addition, different classification methods are investigated. ECG signals are decomposed into eight levels using a Symlet function as a mother Wavelet function with third order. Then, the entropy-based features including fuzzy/approximate/sample/correct conditional entropy as well as other nonlinear features including interquartile range, mean absolute deviation, variance, Poincare plot, and recurrence plot are extracted from WT coefficients. The best features are chosen using the automatic sequential forward feature selection algorithm. In order to assess the introduced method, 95 single-lead ECG recordings are used. The support vector machine classifier having a radial basis function kernel leads to an accuracy of 94.63% (sensitivity: 94.43% and specificity: 94.77%) and 95.71% (sensitivity: 95.83% and specificity: 95.66%) for minute-by-minute and subject-by-subject classifications, respectively. The results show that applying entropy-based features for extracting hidden information of the ECG signals outperforms other available automatic OSA detection methods. The results indicate that a highly accurate OSA detection is attained by just exploiting the single-lead ECG signals. Furthermore, due to the low computational load in the proposed method, it can easily be applied to the home monitoring systems.