Prediction of paroxysmal Atrial Fibrillation: A machine learning based approach using combined feature vector and mixture of expert classification on HRV signal. (October 2018)

• Record Type: Journal Article
• Authors: Ebrahimzadeh, Elias; Kalantari, Maede; Joulani, Mohammadamin; Shahraki, Reza Shahrokhi; Fayaz, Farahnaz; Ahmadi, Fereshteh
• Is Part Of: Computer methods and programs in biomedicine. Volume 165(2018)
• Page Start: 53
• Page End: 67
• Publication Date: 2018-10
• Subjects: Paroxysmal atrial fibrillation -- Heart rate variability -- Feature reduction -- Local subset feature selection -- Mixture of Expert; Medicine -- Computer programs -- Periodicals; Biology -- Computer programs -- Periodicals; Computers -- Periodicals; Medicine -- Periodicals; Médecine -- Logiciels -- Périodiques; Biologie -- Logiciels -- Périodiques; Biology -- Computer programs; Medicine -- Computer programs; Periodicals; Electronic journals; 610.28
• DOI: 10.1016/j.cmpb.2018.07.014
• Languages: English
• ISSNs: 0169-2607

Abstract

Highlights: This research has applied a novel and automatic approach to ensure Local Feature Subset Selection with the assistance of the most rigorous methodologies, which have formerly been developed in previous works of our team, for extracting features from nonlinear, time-frequency and classical processes. Not only will this facilitate increasing the prediction accuracy, but it also will provide us with an opportunity to interpret clinical signs considering the plurality of features. In this work, the most applicable and effective features would subsequently be presented according to the Local Subset feature Selection which is a novel approach of machine learning. The results denote that combining the most informative features extracted from different domains along with performing the local subset feature selection method and using the ME classifier lead to a more accurate predictor of PAF onset. The results indicate the significant capacity of the proposed method for predicting PAF as well as selecting the appropriate processing method any time before the incident. While there appears to be no distinct difference between ECG signals that are leading to PAF and those that are distant from PAF, HRV signal contains critical information within its nonlinear nature. To benefit from time domain and frequency features at the same time, we have made use of the time-frequency analyses, particularly Wigner-Ville transform, which ensure both classic and nonlinear methods are included and applied in a complementary fashionThe early prediction of an unexpected PAF risk in a person experiencing VF is highly significant for timely treatment and increased survival rate. Prediction and prevention of PAF is an area of active investigation, even though considerable challenges persist that limit the efficacy and cost-effectiveness of available methodologies. Consequently, there is still an urgent need for a time effective yet precise procedure to predict PAF in an advanced, automated, and clinically applicable manner. On the other hand, considering the importance of this disease, which could be the first and the last heart condition to be diagnosed in a person, and how it continues to claim millions of people's lives around the world, prediction of PAF has increasingly been regarded as a matter of substantive significance. As effective approaches to PAF prediction, based on which doctors can make informed decisions, are yet to be discovered, this research aims to propose a novel approach to local feature selection with the assistance of the most accurate methodologies, which have formerly been developed in previous works of this team. Experimental results show that there are significant information in HRV signal which can be extracted by the proposed method and used for prediction of PAF although there is no significant difference between ECG of a person, immediately before the onset of PAF and distant from the onset in time domain. This study has proposed a new combinational feature vector which contains much more precious information for prediction of PFA in comparison with previous works. From a clinical perspective, the achieved results of PAF prediction have further strengthened our confidence in enabling timely treatment and increasing the survival rate. Finally, our findings about detection of PFA can help doctors and Treatment centers to be aware of onset before happening to preventing incident and do something that save the life of unhealthy person. Abstract: Background and Objective: Paroxysmal Atrial Fibrillation (PAF) is one of the most common major cardiac arrhythmia. Unless treated timely, PAF might transform into permanent Atrial Fibrillation leading to a high rate of morbidity and mortality. Therefore, increasing attention has been directed towards prediction of PAF, to enable early detection and prevent further progression of the disease. Notwithstanding the pharmacological and electrical treatments, a validated method to predict the onset of PAF is yet to be developed. We aim to address this issue through integrating classical and modern methods. Methods: To increase the predictivity, we have made use of a combination of features extracted through linear, time-frequency, and nonlinear analyses performed on heart rate variability. We then apply a novel approach to local feature selection using meticulous methodologies, developed in our previous works, to reduce the dimensionality of the feature space. Subsequently, the Mixture of Experts classification is employed to ensure a precise decision-making on the output of different processes. In the current study, we analyzed 106 signals from 53 pairs of ECG recordings obtained from the standard database called Atrial Fibrillation Prediction Database (AFPDB). Each pair of data contains one 30-min ECG segment that ends just before the onset of PAF event and another 30-min ECG segment at least 45 min distant from the onset. Results: Combining the features that are extracted using both classical and modern analyses was found to be significantly more effective in predicting the onset of PAF, compared to using either analyses independently. Also, the Mixture of Experts classification yielded more precise class discrimination than other well-known classifiers. The performance of the proposed method was evaluated using the Atrial Fibrillation Prediction Database (AFPDB) which led to sensitivity, specificity, and accuracy of 100%, 95.55%, and 98.21% respectively. Conclusion: Prediction of PAF has been a matter of clinical and theoretical importance. We demonstrated that utilising an optimized combination of — as opposed to being restricted to — linear, time-frequency, and nonlinear features, along with applying the Mixture of Experts, contribute greatly to an early detection of PAF, thus, the proposed method is shown to be superior to those mentioned in similar studies in the literature. Graphicalabstract :