J Med Internet Res. 2020;22(5):e16443.

Detection of Atrial Fibrillation Using a Ring-Type Wearable Device (CardioTracker) and Deep Learning Analysis of Photoplethysmography Signals: Prospective Observational Proof-of-Concept Study.

, , , , , , , , ,

제1저자 권순일, 홍준기 · 교신저자 최의근

PMID 32348254 PMC7273241 DOI 10.2196/16443 IF 8.2 순환기내과

소속 (4)

Seoul National University Hospital, Seoul, Republic of Korea.
Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Republic of Korea.
Sky Labs Inc, Seongnam, Republic of Korea.
Department of Information and Communication Engineering, Hanbat National University, Daejeon, Republic of Korea.

Abstract

BACKGROUND Continuous photoplethysmography (PPG) monitoring with a wearable device may aid the early detection of atrial fibrillation (AF).

OBJECTIVE We aimed to evaluate the diagnostic performance of a ring-type wearable device (CardioTracker, CART), which can detect AF using deep learning analysis of PPG signals.

METHODS Patients with persistent AF who underwent cardioversion were recruited prospectively. We recorded PPG signals at the finger with CART and a conventional pulse oximeter before and after cardioversion over a period of 15 min (each instrument). Cardiologists validated the PPG rhythms with simultaneous single-lead electrocardiography. The PPG data were transmitted to a smartphone wirelessly and analyzed with a deep learning algorithm. We also validated the deep learning algorithm in 20 healthy subjects with sinus rhythm (SR).

RESULTS In 100 study participants, CART generated a total of 13,038 30-s PPG samples (5850 for SR and 7188 for AF). Using the deep learning algorithm, the diagnostic accuracy, sensitivity, specificity, positive-predictive value, and negative-predictive value were 96.9%, 99.0%, 94.3%, 95.6%, and 98.7%, respectively. Although the diagnostic accuracy decreased with shorter sample lengths, the accuracy was maintained at 94.7% with 10-s measurements. For SR, the specificity decreased with higher variability of peak-to-peak intervals. However, for AF, CART maintained consistent sensitivity regardless of variability. Pulse rates had a lower impact on sensitivity than on specificity. The performance of CART was comparable to that of the conventional device when using a proper threshold. External validation showed that 94.99% (16,529/17,400) of the PPG samples from the control group were correctly identified with SR.

CONCLUSIONS A ring-type wearable device with deep learning analysis of PPG signals could accurately diagnose AF without relying on electrocardiography. With this device, continuous monitoring for AF may be promising in high-risk populations.

TRIAL REGISTRATION ClinicalTrials.gov NCT04023188; https://clinicaltrials.gov/ct2/show/NCT04023188.

©Soonil Kwon, Joonki Hong, Eue-Keun Choi, Byunghwan Lee, Changhyun Baik, Euijae Lee, Eui-Rim Jeong, Bon-Kwon Koo, Seil Oh, Yung Yi. Originally published in the Journal of Medical Internet Research (http://www.jmir.org), 21.05.2020.

Keywords

atrial fibrillationdeep learningdiagnosisphotoplethysmographywearable electronic devices

Figures

**Figure 1** Demonstration of photoplethysmography (PPG) monitoring by CardioTracker (CART). CART measures PPG signals at the proximal phalanx and wirelessly transmits the data to the linked smartphone, which can monitor the PPG signals in real-time, and the deep learning algorithm suggests a possible diagnosis.

**Figure 2** Diagnostic performance of CardioTracker (CART) according to the algorithms. CART with the deep learning algorithm achieved the highest results for all diagnostic parameters. (A) ROC curves, (B) Diagnostic parameters, and (C) AUCs according to the algorithms. AUC: area under the curve, CNN: convolutional neural network, NPV: negative-predictive value, PPV: positive-predictive value, ROC: receiver operating characteristic, SN: sensitivity, SP: specificity, SVM, autocorrelation: support vector machine with autocorrelation as a feature, SVM, RMSSD+ShE: support vector machine with root mean square of the successive differences of RR intervals and Shannon entropy as features, SVM, ensemble: support vector machine with all three features.

**Figure 3** The diagnostic performance of CardioTracker according to sample length. In general, longer lengths of photoplethysmography samples had higher diagnostic performances. AUC: area under the curve, NPV: negative-predictive value, PPV: positive-predictive value, SN: sensitivity, SP: specificity.

**Figure 4** The specificity of CardioTracker according to the burden of premature beats. (A) The five-fold cross-validation process with randomization of participants. There was a decreasing trend of specificity according to increasing burden of premature beats. However, the convolutional neural network (CNN) maintained the highest results for most cases. (B) The five-fold cross-validation process with randomization of samples. The CNN improved specificity in especially high burden of premature beats. SVM, autocorrelation: support vector machine with autocorrelation as a feature, SVM, RMSSD+ShE: support vector machine with root mean square of the successive differences of RR intervals and Shannon entropy as features, SVM, ensemble: support vector machine with all three features.

**Figure 5** The sensitivity and specificity of CardioTracker according to the characteristics of samples. (A) and (B) With the deep learning algorithm, there were no definite associations between the sensitivity and peak-to-peak interval variability or the pulse rate. (C) The specificity generally decreased with higher peak-to-peak interval variability. (D) There was generally a U-shape association between specificity and the pulse rate. CNN: convolutional neural network, SVM, autocorrelation: support vector machine with autocorrelation as a feature, SVM, RMSSD+ShE: support vector machine with root mean square of the successive differences of RR intervals and Shannon entropy as features, SVM, ensemble: support vector machine with all three features.

**Figure 6** Visualization of deep learning analyses. The deep learning analyses of CardioTracker are plotted with the t-SNE method. The upper panel: (A) The two clusters of AF and SR were well differentiated from each other, leaving a small overlapped potion. (B), (C), and (D) The overlapped region showed low diagnostic confidence, low pulse rates, and modest peak-to-peak interval variability. The lower panel: typical examples of photoplethysmography samples. AF: atrial fibrillation, SR: sinus rhythm, t-SNE: t-distributed stochastic neighbor embedding.

그림 출처: PubMed Central 공개 논문 (PMC7273241)

Conflict of interest statement

Conflicts of Interest: SK, JH, EL, and SO: None declared. EKC, ERJ, BKK, and YY: Stockholders of Sky Labs Inc, Seongnam, Republic of Korea. BL and CB: Employees of Sky Labs Inc.

그 밖의 정보