In this study, we propose a cuff-less blood pressure (BP) estimation technique based on deep learning for smart wristwatches. Photoplethysmography (PPG) and electrocardiography (ECG) signals are first collected from the sensors installed at a smart wristwatch. Ground truth systolic BP (SBP) and diastolic BP (DBP) measurements are then obtained by a reference device, a mercury sphygmomanometer. In order to estimate the SBP and DBP, we extract feature vectors and reconstruct them through a feature selection process. Next, we design two-stage system of a stacked deep neural network (DNN)-based SBP and DBP estimation models and compare our results to those obtained using estimation techniques in the previously reported algorithms such as the polynomial regression (PR), support vector machine (SVM), artificial neural network (ANN), and deep belief network (DBN)-DNN. In order to verify the proposed algorithm against the conventional algorithms, we quantitatively compare the results in terms of mean error (ME) with standard deviation, mean absolute error (MAE) with standard deviation, Pearson correlation, box plot, and Bland-Altman plot. For this, 110 subjects contributed to the database (DB), each of which is collected three times for 20 seconds. The quantitative errors turn out lower compared to the existing methods, which shows the superiority of our approach. To further enhance the BP estimation performance for each individual user, we devise the personal adaptation algorithm for the BP estimation algorithm that yields better the BP estimates.

In this paper, we propose a deep belief network (DBN)-deep neural network (DNN) with mimic features based on the bootstrap inspired technique to learn the complex nonlinear relationship between the mimic feature vectors obtained from the oscillometry signals and the target blood pressures. Unfortunately, we have two problems in utilizing the DBN-DNN technique to estimate the systolic blood pressure (SBP) and diastolic blood pressure (DBP). First, our set of input feature vectors is very small, which is a fatal drawback to training based on the DBN-DNN technique. Second, the special pre-training phase can also trigger an unstable estimation, because there are still a lot of random initialized assigns, such as the training data set, weights, and biases. For these reasons, we employ the bootstrap-inspired technique as a fusion ensemble estimator based on the DBN-DNN-based regression model, which is used to create the mimic features to estimate the SBP and DBP. Our DBN-DNN-based ensemble regression estimator provides a lower standard deviation of error, mean error, and mean absolute error for the SBP and DBP as compared with those of the conventional methods.

Polysomnography (PSG) is considered as the gold standard for determining sleep stages, but due to the obtrusiveness of its sensor attachments, sleep stage classification algorithms using noninvasive sensors have been developed throughout the years. However, the previous studies have not yet been proven reliable. In addition, most of the products are designed for healthy customers rather than for patients with sleep disorder. We present a novel approach to classify sleep stages via low cost and noncontact multi-modal sensor fusion, which extracts sleep-related vital signals from radar signals and a sound-based context-awareness technique. This work is uniquely designed based on the PSG data of sleep disorder patients, which were received and certified by professionals at Hanyang University Hospital. The proposed algorithm further incorporates medical/statistical knowledge to determine personal-adjusted thresholds and devise post-processing. The efficiency of the proposed algorithm is highlighted by contrasting sleep stage classification performance between single sensor and sensor-fusion algorithms. To validate the possibility of commercializing this work, the classification results of this algorithm were compared with the commercialized sleep monitoring device, ResMed S+. The proposed algorithm was investigated with random patients following PSG examination, and results show a promising novel approach for determining sleep stages in a low cost and unobtrusive manner.