Weighted Random Forests to Improve Arrhythmia Classification.

Construction of an ensemble model is a process of combining many diverse base predictive learners. It arises questions of how to weight each model and how to tune the parameters of the weighting process. The most straightforward approach is simply to average the base models. However, numerous studies have shown that a weighted ensemble can provide superior prediction results to a simple average of models. The main goals of this article are to propose a new weighting algorithm applicable for each tree in the Random Forest model and the comprehensive examination of the optimal parameter tuning. Importantly, the approach is motivated by its flexibility, good performance, stability, and resistance to overfitting. The proposed scheme is examined and evaluated on the Physionet/Computing in Cardiology Challenge 2015 data set. It consists of signals (electrocardiograms and pulsatory waveforms) from intensive care patients which triggered an alarm for five cardiac arrhythmia types (Asystole, Bradycardia, Tachycardia, Ventricular Tachycardia, and Ventricular Fultter/Fibrillation). The classification problem regards whether the alarm should or should not have been generated. It was proved that the proposed weighting approach improved classification accuracy for the three most challenging out of the five investigated arrhythmias comparing to the standard Random Forest model.

Reducing False Arrhythmia Alarms Using Different Methods of Probability and Class Assignment in Random Forest Learning Methods.

The literature indicates that 90% of clinical alarms in intensive care units might be false. This high percentage negatively impacts both patients and clinical staff. In patients, false alarms significantly increase stress levels, which is especially dangerous for cardiac patients. In clinical staff, alarm overload might lead to desensitization and could result in true alarms being ignored. In this work, we applied the random forest method to reduce false arrhythmia alarms and specifically explored different methods of probability and class assignment, as these affect the classification accuracy of the ensemble classifiers. Due to the complex nature of the problem, i.e., five types of arrhythmia and several methods to determine probability and the alarm class, a synthetic measure based on the ranks was proposed. The novelty of this contribution is the design of a synthetic measure that helps to leverage classification results in an ensemble model that indicates a decision path leading to the best result in terms of the area under the curve (AUC) measure or the global accuracy (score). The results of the research are promising. The best performance in terms of the AUC was 100% accuracy for extreme tachycardia, whereas the poorest results were for ventricular tachycardia at 87%. Similarly, in terms of the accuracy, the best results were observed for extreme tachycardia (91%), whereas ventricular tachycardia alarms were the most difficult to detect, with an accuracy of only 51%.

RS slope detection algorithm for extraction of heart rate from noisy, multimodal recordings.

Current gold-standard algorithms for heart beat detection do not work properly in the case of high noise levels and do not make use of multichannel data collected by modern patient monitors. The main idea behind the method presented in this paper is to detect the most prominent part of the QRS complex, i.e. the RS slope. We localize the RS slope based on the consistency of its characteristics, i.e. adequate, automatically determined amplitude and duration. It is a very simple and non-standard, yet very effective, solution. Minor data pre-processing and parameter adaptations make our algorithm fast and noise-resistant. As one of a few algorithms in the PhysioNet/Computing in Cardiology Challenge 2014, our algorithm uses more than two channels (i.e. ECG, BP, EEG, EOG and EMG). Simple fundamental working rules make the algorithm universal: it is able to work on all of these channels with no or only little changes. The final result of our algorithm in phase III of the Challenge was 86.38 (88.07 for a 200 record test set), which gave us fourth place. Our algorithm shows that current standards for heart beat detection could be improved significantly by taking a multichannel approach. This is an open-source algorithm available through the PhysioNet library.