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

Abstract

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%.

Weighted Random Forests to Improve Arrhythmia Classification

“…Distribution of the recordings among the investigated five types of arrhythmias and whether the alarm should or should not have been generated are: Asystole (No-100, Yes-22); Bradycardia (No-43, Yes-46); Tachycardia (No-9, Yes-131); Ventricular Tachycardia (No-252, Yes-89); Ventricular Fibrillation or Flutter (No-52, Yes-6). The signals provided were already pre-filtered with multiple notch filters and finite impulse response (FIR) band pass filter (0.05-40 Hz) [18,19].…”

“…As mentioned in Section 2, to diagnose Asystole, Bradycardia, and Tachycardia it is critical to properly locate consecutive heart beats. Hence, the first step to create features was the detection of QRS complexes in the ECG signal, performed by a low-complexity R-peak detector as described in [18,51]. At the same time, an open source wabp algorithm [19] was used to locate the beats in pulsatile waveforms provided (ABP, PLETH).…”

“…To evaluate the proposed weighting approach, numerical experiments on five data sets regarding arrhythmia classification have been conducted. To demonstrate the generalizability of the proposed scheme we compare achieved results to the results obtained in our previous study (standard version of Random Forest) [18] examining data provided by organizers of the Physionet/Computing in Cardiology Challenge [19]. During the challenge, participants were provided with 750 signals registered 5 min before the alarm generation and information about what type of arrhythmia caused the alarm.…”

“…The alarms were triggered by five types of arrhythmia: Asystole, Bradycardia, Tachycardia, Ventricular Fibrillation or Flutter, and Ventricular Tachycardia. All the signals were analyzed by expert annotators and labelled as true or false [18,19].…”

Single-modal and multi-modal false arrhythmia alarm reduction using attention-based convolutional and recurrent neural networks

“…Then, several machine learning algorithms are evaluated using the extracted features to detect false alarms. In [20], a random forest technique is applied to reduce false alarms using different methods of probability and class assignments. Lehman et al [21] adopted a supervised denoising autoencoder (SDAE) to identify false alarms in Ventricular Tachycardia using features of interest extracted by FFT.…”