We develop an algorithm that accurately detects Atrial Fibrillation (AF) episodes from photoplethysmograms (PPG) recorded in ambulatory free-living conditions. We collect and annotate a dataset containing more than 4000 hours of PPG recorded from a wrist-worn device. Using a 50-layer convolutional neural network, we achieve a test AUC of 95% and show robustness to motion artifacts inherent to PPG signals. Continuous and accurate detection of AF from PPG has the potential to transform consumer wearable devices into clinically useful medical monitoring tools. * indicates equal contribution arXiv:1811.07774v2 [physics.med-ph]

No abstract

Abstract-We used kernel density estimation (KDE) methods to build a priori probability density functions (pdfs) for the vector of features that are used to classify unexploded ordnance items given electromagnetic-induction sensor data. This a priori information is then used to develop a new suite of estimation and classification algorithms. As opposed to the commonly used maximumlikelihood parameter estimation methods, here we employ a maximum a posteriori (MAP) estimation algorithm that makes use of KDE-generated pdfs. Similarly, we use KDE priors to develop a suite of classification schemes operating in both "feature" space as well as "signal/data" space. In terms of feature-based methods, we construct a support vector machine classifier and its extension to support M -ary classification. The KDE pdfs are also used to synthesize a MAP feature-based classifier. To address the numerical challenges associated with the optimal data-space Bayesian classifier, we have used several approximation techniques, including Laplacian approximation and generalized likelihood ratio tests employing the priors. Using both simulations and real field data, we observe a significant improvement in classification performance due to the use of the KDE-based prior models.

We used kernel density estimation to build a-priori probability distributions on the vector of features used to characterize unexploded ordnance from electromagnetic induction sensor data. This priori information is then used in a Bayesian framework to develop a new suite of estimation and classification algorithms. Based on this prior information several classification algorithms are developed in feature and signal space. Results using real field data show more robust estimation and significant improvement in classification performance for signal space classifiers comparing to conventional Gaussian approximation to the density of the features.

No abstract