A ballistocardiograph records the mechanical activity of the heart. We present a novel algorithm for the detection of individual heart beats and beat-to-beat interval lengths in ballistocardiograms (BCGs) from healthy subjects. An automatic training step based on unsupervised learning techniques is used to extract the shape of a single heart beat from the BCG. Using the learned parameters, the occurrence of individual heart beats in the signal is detected. A final refinement step improves the accuracy of the estimated beat-to-beat interval lengths. Compared to many existing algorithms, the new approach offers heart rate estimates on a beat-to-beat basis. The agreement of the proposed algorithm with an ECG reference has been evaluated. A relative beat-to-beat interval error of 1.79% with a coverage of 95.94% was achieved on recordings from 16 subjects.

Abstract. Reliable and accurate estimation of instantaneous frequencies of physiological rhythms, such as heart rate, is critical for many healthcare applications. Robust estimation is especially challenging when novel unobtrusive sensors are used for continuous health monitoring in uncontrolled environments, because these sensors can create significant amounts of potentially unreliable data. We propose a new flexible algorithm for the robust estimation of local (beat-to-beat) intervals from ballistocardiograms (BCGs) recorded by an unobtrusive bed-mounted sensor. This sensor allows the measurement of motions of the body which are caused by cardiac activity.Our method requires neither a training phase nor any prior knowledge about the morphology of the heart beats in the analyzed waveforms. Instead, three short-time estimators are combined using a Bayesian approach to continuously estimate the interbeat intervals. We have validated our method on over-night BCG recordings from 33 subjects (8 normal, 25 insomniacs). On this dataset, containing approximately 1 million heart beats, our method achieved a mean beat-to-beat interval error of 0.94 % with a coverage of 70.96 %.

Monitoring vital signs through unobtrusive means is a goal which has attracted a lot of attention in the past decade. This review provides a systematic and comprehensive review over the current state of the field of ambient and unobtrusive cardiorespiratory monitoring. To this end, nine different sensing modalities which have been in the focus of current research activities are covered: capacitive electrocardiography, seismo- and ballistocardiography, reflective photoplethysmography (PPG) and PPG imaging, thermography, methods relying on laser or radar for distance-based measurements, video motion analysis, as well as methods using high-frequency electromagnetic fields. Current trends in these subfields are reviewed. Moreover, we systematically analyze similarities and differences between these methods with respect to the physiological and physical effects they sense as well as the resulting implications. Finally, future research trends for the field as a whole are identified.

We present a study on the feasibility of the automatic detection of atrial fibrillation (AF) from cardiac vibration signals (ballistocardiograms/BCGs) recorded by unobtrusive bedmounted sensors. The proposed system is intended as a screening and monitoring tool in home-healthcare applications and not as a replacement for ECG-based methods used in clinical environments. Based on BCG data recorded in a study with 10 AF patients, we evaluate and rank seven popular machine learning algorithms (naive Bayes, linear and quadratic discriminant analysis, support vector machines, random forests as well as bagged and boosted trees) for their performance in separating 30 s long BCG epochs into one of three classes: sinus rhythm, atrial fibrillation, and artifact. For each algorithm, feature subsets of a set of statistical time-frequency-domain and time-domain features were selected based on the mutual information between features and class labels as well as first- and second-order interactions among features. The classifiers were evaluated on a set of 856 epochs by means of 10-fold cross-validation. The best algorithm (random forests) achieved a Matthews correlation coefficient, mean sensitivity, and mean specificity of 0.921, 0.938, and 0.982, respectively.