An end-edge coordinated multiview deep learning method improved atrial fibrillation detection accuracy by 6.62% and F1 score by 11.21% on average compared with conventional methods.
Does a multiview deep learning method improve atrial fibrillation detection accuracy compared to conventional methods?
A novel multiview deep learning approach utilizing signal, spectrogram, and Lorenz R-R scatter plots significantly improves the accuracy of automated atrial fibrillation detection.
Estimación del efecto: +6.62% accuracy, +11.21% F1 score
Atrial Fibrillation (AF), one of the most common arrhythmias, increases the risk of heart failure, dementia, and stroke. Conventional time-efficient AF detection methods extract the essential features (e.g., heart rate stability and P-wave evidence) based on the medical knowledge to conduct the triage. Deep learning enables a new way of AF detection by training the deep model with only labeled data, significantly reducing the expertise required. However, most existing deep learning methods utilize only one view of signal or spectrogram and neglect the information from the medical perspective, which could result in relatively low detection accuracy. Moreover, most multiview methods cannot be directly deployed on the cardiac monitor due to the network size and the algorithm complexity. To address the challenges, we propose a multiview deep learning method and an end-edge coordinated system for accurate and time-efficient AF detection. To be more specific, the multiview method utilizes standard and medically related views, including signal, spectrogram, and Lorenz R-R scatter plot, which fully explore the information of the heart rhythm. We further devise an end-edge coordinated inference system to achieve time-efficient and accurate AF detection on the resource-constrained cardiac monitor. Experimental results show that our method improves the accuracy and the F1 score by 6.62% and 11.21% on average when compared with the conventional method.
Zhou et al. (Thu,) conducted a other in Atrial Fibrillation. End-edge coordinated multiview deep learning method vs. Conventional method was evaluated on Accuracy and F1 score (+6.62% accuracy, +11.21% F1 score). An end-edge coordinated multiview deep learning method improved atrial fibrillation detection accuracy by 6.62% and F1 score by 11.21% on average compared with conventional methods.