A Novel Method for Calibration-Based Cuff-Less Blood Pressure Estimation.

Cuff-less blood pressure estimation technology is useful for cardiovascular disease monitoring. However, without calibration, cuff-less blood pressure estimation is hard to achieve clinical acceptable performance. The traditional methods are always calibrated with retraining. With the increases of the parameters number, the cost of model retraining increases several times. So we propose a novel blood pressure estimation method, which can be calibrated with reference inputs rather than with retraining. The experiment results suggest that the method we proposed can achieve clinical performance (SBP:-0.004 ± 5.869 mmHg, DBP:-0.004±4.511 mmHg) with low calibration cost.

A Novel Detection Method of Bundle Branch Block from Multi-lead ECG.

Bundle branch block (BBB) is a common conduction block disease and can be diagnosed using electrocardiogram (ECG) signal in clinical practice. In this paper, a novel method was proposed to detect two types of BBB: right BBB (RBBB) and left BBB (LBBB) based on the combination of deep features and several kinds of expert features. We evaluated the proposed method on the MIT-BIH Arrhythmia database (AR) and China Physiological Signal Challenge 2018 database (CPSC). The proposed method achieved an accuracy of 99.96% (AR) in the class-oriented evaluation and an accuracy of 98.76% (AR) and 97.88% (CPSC) in the subject-oriented evaluation, better than the baseline methods. Experimental results show that our method would be a good choice for the detection of the BBB.

Deep Learning Based Patient-Specific Classification of Arrhythmia on ECG signal.

The classification of the heartbeat type is an essential function in the automatical electrocardiogram (ECG) analysis algorithm. The guideline of the ANSI/AAMI EC57 defined five types of heartbeat: non-ectopic or paced beat (N), supraventricular ectopic beat (S), ventricular ectopic beat (V), fusion of a ventricular and normal beat (F), pace beat or fusion of a paced and a normal or beat that cannot be classified (Q). In the work, a deep neural network based method was proposed to classify these five types of heartbeat. After removing the noise from ECG signals by a low-pass filter, the two-lead heartbeat segments with 2-s length were generated on the filtered signals, and classified by an adaptive ResNet model. The proposed method was evaluated on the MIT-BIH Arrhythmia Database with the patient-specific pattern. The overall accuracy was 98.6% and sensitivity of N, S, V, F were 99.4%, 85.4%, 96.6%, 90.6% respectively. Experimental results show that the proposed method achieved a good performance, and would be useful in the clinic practice.

An Electrocardiogram Delineator via Deep Segmentation Network.

Electrocardiogram (ECG) delineation is a process to detect multiple characteristic points, which contain critical diagnostic information about cardiac diseases. We treat the ECG delineation task as an one-dimensional segmentation problem, and propose a novel end-to-end deep learning method to segment sections of ECG signal. Our neural network consists of two parts: a segmentation network composed of multiple 1D Convolutional Neural Networks (CNN) and a postprocessing network composed of a sequential Conditional Random Field (CRF). Our method is trained and validated on QT database. The experimental results show that our method yields competitive overall performance compared with other state-of-the-art works and outperform them on onset of the P wave and offset of the T wave.

Novel Deep Convolutional Neural Network for Cuff-less Blood Pressure Measurement Using ECG and PPG Signals.

Cuff-less blood pressure (BP) is a potential method for BP monitoring because it is undisturbed and continuous monitoring. Existing cuff-less estimation methods are easily influenced by signal noise and non-ideal signal morphology. In this study we propose a novel well-designed Convolutional Neural Network (CNN) model named Deep-BP for BP estimation task. The structure of Deep-BP can help to capture more underlying data features associated with BP than handcrafted features, thus improving the robustness and estimation accuracy. We carry out experiments with and without calibration procedure in training stage to evaluate the performance of new method in different application scenarios. The experiment results show that the Deep-BP model achieves high accuracy and outperforms existing methods, in the experiments both with and without calibration.

Optic Disc Segmentation from Retinal Fundus Images via Deep Object Detection Networks.

Accurate optic disc (OD) segmentation is a fundamental step in computer-aided ocular disease diagnosis. In this paper, we propose a new pipeline to segment OD from retinal fundus images based on deep object detection networks. The fundus image segmentation problem is redefined as a relatively more straightforward object detection task. This then allows us to determine the OD boundary simply by transforming the predicted bounding box into a vertical and non-rotated ellipse. Using Faster R-CNN as the object detector, our method achieves state-of-the-art OD segmentation results on ORIGA dataset, outperforming existing methods in this field.