Reducing dimensionality in a database of sleep EEG arousals

Multi-task learning for arousal and sleep stage detection using fully convolutional networks

Objective.Sleep is a critical physiological process that plays a vital role in maintaining physical and mental health. Accurate detection of arousals and sleep stages is essential for the diagnosis of sleep disorders, as frequent and excessive occurrences of arousals disrupt sleep stage patterns and lead to poor sleep quality, negatively impacting physical and mental health. Polysomnography is a traditional method for arousal and sleep stage detection that is time-consuming and prone to high variability among experts.Approach. In this paper, we propose a novel multi-task learning approach for arousal and sleep stage detection using fully convolutional neural networks. Our model, FullSleepNet, accepts a full-night single-channel EEG signal as input and produces segmentation masks for arousal and sleep stage labels. FullSleepNet comprises four modules: a convolutional module to extract local features, a recurrent module to capture long-range dependencies, an attention mechanism to focus on relevant parts of the input, and a segmentation module to output final predictions.Main results.By unifying the two interrelated tasks as segmentation problems and employing a multi-task learning approach, FullSleepNet achieves state-of-the-art performance for arousal detection with an area under the precision-recall curve of 0.70 on Sleep Heart Health Study and Multi-Ethnic Study of Atherosclerosis datasets. For sleep stage classification, FullSleepNet obtains comparable performance on both datasets, achieving an accuracy of 0.88 and an F1-score of 0.80 on the former and an accuracy of 0.83 and an F1-score of 0.76 on the latter.Significance. Our results demonstrate that FullSleepNet offers improved practicality, efficiency, and accuracy for the detection of arousal and classification of sleep stages using raw EEG signals as input.

A Review of Feature Selection Methods for Machine Learning-Based Disease Risk Prediction

Machine learning has shown utility in detecting patterns within large, unstructured, and complex datasets. One of the promising applications of machine learning is in precision medicine, where disease risk is predicted using patient genetic data. However, creating an accurate prediction model based on genotype data remains challenging due to the so-called “curse of dimensionality” (i.e., extensively larger number of features compared to the number of samples). Therefore, the generalizability of machine learning models benefits from feature selection, which aims to extract only the most “informative” features and remove noisy “non-informative,” irrelevant and redundant features. In this article, we provide a general overview of the different feature selection methods, their advantages, disadvantages, and use cases, focusing on the detection of relevant features (i.e., SNPs) for disease risk prediction.

Grab'Em: A Novel Graph-Based Method for Combining Feature Subset Selectors

Feature selection is one of the most frequent tasks in data mining applications. Its ability to remove useless and redundant features improves the classification performance and gains knowledge about a given problem makes feature selection a common first step in data mining. In many feature selection applications, we need to combine the results of different feature selection processes. The two most common scenarios are the ensembles of feature selectors and the scaling up of feature selection methods using a data division approach. The standard procedure is to store the number of times every feature has been selected as a vote for the feature and then evaluate different selection thresholds with a certain criterion to obtain the final subset of selected features. However, this method is suboptimal as the relationships of the features are not considered in the voting process. Two redundant features may be selected a similar number of times due to the different sets of instances used each time. Thus, a voting scheme would tend to select both of them. In this article, we present a new approach: instead of using only the number of times a feature has been selected, the approach considers how many times the features have been selected together by a feature selection algorithm. The proposal is based on constructing an undirected graph where the vertices are the features, and the edges count the number of times every pair of instances has been selected together. This graph is used to select the best subset of features, avoiding the redundancy introduced by the voting scheme. The proposal improves the results of the standard voting scheme in both ensembles of feature selectors and data division methods for scaling up feature selection.

An ensemble-based feature selection framework to select risk factors of childhood obesity for policy decision making

BACKGROUND

The increasing prevalence of childhood obesity makes it essential to study the risk factors with a sample representative of the population covering more health topics for better preventive policies and interventions. It is aimed to develop an ensemble feature selection framework for large-scale data to identify risk factors of childhood obesity with good interpretability and clinical relevance.

METHODS

We analyzed the data collected from 426,813 children under 18 during 2000-2019. A BMI above the 90th percentile for the children of the same age and gender was defined as overweight. An ensemble feature selection framework, Bagging-based Feature Selection framework integrating MapReduce (BFSMR), was proposed to identify risk factors. The framework comprises 5 models (filter with mutual information/SVM-RFE/Lasso/Ridge/Random Forest) from filter, wrapper, and embedded feature selection methods. Each feature selection model identified 10 variables based on variable importance. Considering accuracy, F-score, and model characteristics, the models were classified into 3 levels with different weights: Lasso/Ridge, Filter/SVM-RFE, and Random Forest. The voting strategy was applied to aggregate the selected features, with both feature weights and model weights taken into consideration. We compared our voting strategy with another two for selecting top-ranked features in terms of 6 dimensions of interpretability.

RESULTS

Our method performed the best to select the features with good interpretability and clinical relevance. The top 10 features selected by BFSMR are age, sex, birth year, breastfeeding type, smoking habit and diet-related knowledge of both children and mothers, exercise, and Mother's systolic blood pressure.

CONCLUSION

Our framework provides a solution for identifying a diverse and interpretable feature set without model bias from large-scale data, which can help identify risk factors of childhood obesity and potentially some other diseases for future interventions or policies.

Identification of Patterns of Cognitive Impairment for Early Detection of Dementia

Early detection of dementia is crucial to devise effective interventions. Comprehensive cognitive tests, while being the most accurate means of diagnosis, are long and tedious, thus limiting their applicability to a large population, especially when periodic assessments are needed. The problem is compounded by the fact that people have differing patterns of cognitive impairment as they progress to different forms of dementia. This paper presents a novel scheme by which individual-specific patterns of impairment can be identified and used to devise personalized tests for periodic follow-up. Patterns of cognitive impairment are initially learned from a population cluster of combined normals and cognitively impaired subjects, using a set of standardized cognitive tests. Impairment patterns in the population are identified using a 2-step procedure involving an ensemble wrapper feature selection followed by cluster identification and analysis. These patterns have been shown to correspond to clinically accepted variants of Mild Cognitive Impairment (MCI), a prodrome of dementia. The learned clusters of patterns can subsequently be used to identify the most likely route of cognitive impairment, even for pre-symptomatic and apparently normal people. Baseline data of 24,000 subjects from the NACC database was used for the study.

Large-scale validation of an automatic EEG arousal detection algorithm using different heterogeneous databases

OBJECTIVE

To assess the validity of an automatic EEG arousal detection algorithm using large patient samples and different heterogeneous databases.

METHODS

Automatic scorings were confronted with results from human expert scorers on a total of 2768 full-night PSG recordings obtained from two different databases. Of them, 472 recordings were obtained during a clinical routine at our sleep center and were subdivided into two subgroups of 220 (HMC-S) and 252 (HMC-M) recordings each, according to the procedure followed by the clinical expert during the visual review (semi-automatic or purely manual, respectively). In addition, 2296 recordings from the public SHHS-2 database were evaluated against the respective manual expert scorings.

RESULTS

Event-by-event epoch-based validation resulted in an overall Cohen's kappa agreement of κ = 0.600 (HMC-S), 0.559 (HMC-M), and 0.573 (SHHS2). Estimated inter-scorer variability on the datasets was, respectively, κ = 0.594, 0.561 and 0.543. Analyses of the corresponding Arousal Index scores showed associated automatic-human repeatability indices ranges of 0.693-0.771 (HMC-S), 0.646-0.791 (HMC-M), and 0.759-0.791 (SHHS2).

CONCLUSIONS

Large-scale validation of our automatic EEG arousal detector on different databases has shown robust performance and good generalization results comparable to the expected levels of human agreement. Special emphasis was put on reproducibility of the results; implementation of our method has been made available online as open source code.

A simple and robust method for the automatic scoring of EEG arousals in polysomnographic recordings

BACKGROUND

Clinical diagnosis of sleep disorders relies on the polysomnographic test to examine the neurophysiological markers of the sleep process. In this test, the recording of the electroencephalographic activity and the submental electromyogram is the source of the analysis for the detection of electroencephalographic arousals. The identification of these events is important for the evaluation of the sleep continuity because they cause the fragmentation of the normal sleep process. This work proposes a new technique for the automatic detection of arousals in polysomnographic recordings, presenting a non-computationally complex method with the idea of providing an easy integration with other algorithms.

METHODS

The proposed algorithm combines different well-known signal analysis solutions to identify relevant arousal patterns with special emphasis on robustness and artifacts tolerance. It is a multistage method that after obtaining an initial set of events, improves the detection finding common EEG arousal patterns. Finally, false positives are discarded after examining each candidate within the context of clinical definitions.

RESULTS

22 polysomnographic recordings from real patients were used to validate the method. The results obtained were encouraging, achieving a precision value of 0.86 and a F₁ score value of 0.79. When compared with the gold standard, the method achieves a substantial agreement (Kappa coefficient of 0.78), with an almost perfect agreement with ten recordings.

CONCLUSIONS

The algorithm designed achieved encouraging results and shows robust behavior in presence of signal artifacts. Its low-coupled design allows its implementation on different development platforms, and an easy combination with other methods.

Computer-assisted diagnosis of the Sleep Apnea-Hypopnea Syndrome: An overview of different approaches

Automatic diagnosis of the Sleep Apnea-Hypopnea Syndrome (SAHS) has become an important area of research due to the growing interest in the field of sleep medicine, and the costs associated to its manual diagnosis. The increment and heterogeneity of the different techniques, however, makes somewhat difficult to adequately follow recent developments. In this paper an overview within the area of computer-assisted diagnosis of SAHS has been performed. This overview of the different methods is presented together with a critical discussion of the current state-of-the-art.

Computer-Assisted Diagnosis of the Sleep Apnea-Hypopnea Syndrome: A Review

Automatic diagnosis of the Sleep Apnea-Hypopnea Syndrome (SAHS) has become an important area of research due to the growing interest in the field of sleep medicine and the costs associated with its manual diagnosis. The increment and heterogeneity of the different techniques, however, make it somewhat difficult to adequately follow the recent developments. A literature review within the area of computer-assisted diagnosis of SAHS has been performed comprising the last 15 years of research in the field. Screening approaches, methods for the detection and classification of respiratory events, comprehensive diagnostic systems, and an outline of current commercial approaches are reviewed. An overview of the different methods is presented together with validation analysis and critical discussion of the current state of the art.