1. Introduction
One of the major diseases associated with the elderly is a stroke and is considered as the second leading cause of death with the third most common cause of disability-adjusted life years (DALYs) [1,2]. Stroke is a medical emergency condition, and prompt treatment is crucial also; earlier action can reduce brain damage and other complications. The risk of stroke increases significantly for elderly adults [3,4]. According to the World Health Organization (WHO), by 2050, the world’s population of those aged 60 years and beyond is projected to reach two billion, up from 900 million in 2015. In Korea, stroke is a major health burden that will substantially increase in the near future, and Korea is becoming the most rapidly aging society in the world [5]. A study based on the stroke incidence rate of some countries reported that 75–89% of strokes occur in individuals aged >65 years; of those, 50% occur in people aged ≥70 years and 25% in those who are aged >85 years [6]. As of 2013, the global mortality and disability rate caused by stroke were nearly 6.5 million and 113 million, respectively, whereas the mortality rate is higher in Asia than in Europe and the Americas [7]. Another study reported that the stroke mortality rate of Korea sharply increased, particularly after the age of 70 years [2]. Approximately, every year, 105,000 people experience a new or recurrent stroke, and more than 26,000 people die due to stroke, which indicates that, every five min, a stroke attacks someone, and in every 20 min, a stroke kills someone in Korea. This results in a substantial economic burden to Korea, with the total nationwide cost for stroke care nearly 3.3 billion US dollars in 2005 [1]. The duration of hospital stay and medical expenditure is higher in stroke patients than patients with other chronic diseases [8]. Therefore, an early diagnosis of stroke can enable us to save the lives of people, and the research on stroke patients is also very important for the effective utilization of medical resources.Conventionally, stroke has been diagnosed by various methods, such as computed tomography (CT), magnetic resonance imaging (MRI), a blood test and electrocardiogram (ECG) signals. ECG has always been considered as a popular measurement scheme that is comparatively low-cost in screening and diagnosing various diseases [9]. It has a variety of applications in health monitoring and auxiliary diagnosis and plays a valuable complementary technique for stroke diagnostics [10]. ECG has been mainly used for stroke detection, which helps to determine the cause of stroke. Since the investigation of ECG signals is noninvasive by placing the electrodes on the skin, this helps in the accurate detection of abnormalities. Therefore, a baseline ECG can be useful for detecting cardiac abnormalities in acute stroke patients [11].In medicine, machine learning has great hope in predicting the disease and assisting doctors in diagnosis, using data collected by wearable sensors and smartphones. The use-case of the machine-learning model is both a diagnosis or prognosis in which the diagnostic models can be used for new subjects, and the model developed for prognosis can predict a given subject’s future clinical state [12]. The machine-learning and deep-learning techniques such as logistic regression, random forest and deep neural networks are used for predicting the presence of stroke disease with various related attributes [13,14]. The stroke risk classification techniques are developed by using logistic regression, Naïve Bayesian, Bayesian network, decision tree, neural network, random forest, bagged decision tree, voting and boosting models [15]. Additionally, the machine-learning algorithms are capable of identifying the features that are highly related to stroke occurrence efficiently from the huge set of features [16].Several studies focused on patient classification based on the overall behavior of the ECG to diagnose specific diseases [17]. The extraction of features from the ECG signal is a key step for ECG recognition, as it allows to greatly enhance and extract the characteristics of the signal, and those features can be fed into the conventional machine-learning models to perform the classification [18]. The integration of classification techniques in the clinical setup majorly requires the detection of ECG abnormalities in real time to be used in the hospital environment at the bedside or on wearable devices [17]. In general, ECG abnormalities are frequently seen in patients with a stroke. A study revealed that any ECG abnormality is a highly ranked factor, and it may be possible that all ECG abnormalities are more indicative of stroke than just atrial fibrillation [16]. Therefore, the analysis of ECG features and the detection of abnormalities from the ECG signal are significant tasks for the diagnosis of stroke disease. With the consideration of the importance of ECG changes in stroke [19], our study aimed to analyze the ECG features for the classification of elderly post-stroke patients and control subjects with a stroke diagnostic approach. Firstly, we extracted the features from the denoised ECG signals of the study participants. Secondly, we performed the analyses on the features such as heart rate variability (HRV) indices of time and spectral domains, fiducial features and statistical and impulsive metrics variables that have not been commonly used. Thirdly, we employed the filter-based feature selection approach for selecting the input feature subset for the classifiers. Finally, we intended to develop a classification model for diagnosing stroke disease based on machine-learning techniques such as k-nearest neighbor (KNN), support vector machine (SVM), Naïve Bayes, random forest and logistic regression. Furthermore, our study aimed to investigate the autonomic dysfunction and the potential risk factors in relation to stroke.
Read more Kapal Bhati Pranayama - Skull Shining Breathing Technique