A Patient-Specific Methodology for Prediction of Paroxysmal Atrial Fibrillation Onset

Abstract

In spite of the progress in management of Atrial Fibrillation (AF), this arrhythmia is one of the major causes of stroke and heart failure. The progression of this pathology from a silent paroxysmal form (PAF) into a sustained AF can be prevented by predicting the onset of PAF episodes. Moreover, since AF is caused by heterogeneous mechanisms in different patients, as we demonstrate in this paper, a patient-specific approach offers a promising solution. In this work, we consider two ECG recordings, one close to PAF onset and one far away from any PAF episode. For each patient, we extract two 5-minute ECG segments approximately 20 minutes apart. Next, we train a linear Support Vector Machine (SVM) classifier using patientspecific sets of time-and amplitude-domain features. In particular, we consider the P-waves and the QRS complexes in short windows of 5 consecutive heart beats. Finally, we validate the method on the PAF Prediction Challenge (2001) PhysioNet database predicting the onset with an F1 score of 97.1%, sensitivity of 96.2% and specificity of 98.1%.

Online Obstructive Sleep Apnea Detection on Medical Wearable Sensors

“…r We propose an online Obstructive Sleep Apnea detection technique compatible with cardiac-monitoring as developed by Sopic et al, as well as De Giovanni et al [43]- [45], with a time-complexity of O(n), which is the theoretical lower bound. This is achieved by developing our own efficient outlier removal and through performing sleep apnea assessment in time-domain, which removes the need for computationally expensive frequency-domain analysis (see Section IV).…”

Real-Time Personalized Atrial Fibrillation Prediction on Multi-Core Wearable Sensors

“…Additionally, we need one excerpt of normal sinus ECG signal at least 45 min far from any event (i.e., "Far from PAF"). We choose a training window of at least 350 R peaks for both excerpts, which corresponds to 3-9 min considering a heart rate varying from 40 BPM to 110 BPM, as done in [26]. Moreover, to avoid overfitting and ensure robustness and generalization of the training model we perform a learning curve analysis [27] on the number of training samples.…”

“…The R peak detection step generates a set of adaptive hysteresis thresholds to isolate the highly dominant peaks and performs a subsequent check on their widths to eliminate false positives. Second, we apply a method described in our previous work [26] to detect the onset and offset of the P wave by comparing it with a triangular wave starting at the P peak and finishing at the isoelectric line. Third, we delineate the S wave as the minimum point after the R peak, by considering the standard physiological duration of the QRS complex.…”

“…Moreover, Compute mean F-score on folds 13: end for 14: Choose group with best mean F-score 15: return group we select a personalized combination of features in a specific amount of consecutive beats that defines a small window of processing. In this way we analyze sudden events occurring within the window, as shown in our previous work [26]. Additionally, we include an overlapping window depending on the patient.…”

“…In order to draw a comparison with our method, we report two key cases of the recent literature [21], [22], describing offline methodologies, which include inter-patient variability and achieve higher accuracy compared to other methods [23]- [25]. Finally, we compare the accuracy results with our previous work on offline PAF prediction [26].…”