Estimation of atrial fibrillation from lead-I ECGs: Comparison with cardiologists and machine learning model (CurAlive), a clinical validation study

15 Apr 2021  ·  N. Korucuk, C. Polat, E. S. Gunduz, O. Karaman, V. Tosun, M. Onac, N. Yildirim, Y. Cete, K. Polat ·

Electrocardiogram recognition of cardiac arrhythmias is critical for cardiac abnormality diagnosis. Because of their strong prediction characteristics, artificial neural networks are the preferred method in medical diagnosis systems. This study presents a method to detect atrial fibrillation with lead-I ECGs using artificial intelligence. The aim of the study is to compare the accuracy of the diagnoses estimated by cardiologists and artificial intelligence over lead-I ECGs using 12-lead ECGs as references. To evaluate the performance of the proposed model, dataset were collected from China Physiological Signal Challenge 2018. In the study, diagnoses were examined in three groups as normal sinus rhythm, atrial fibrillation and OTHER. All rhythm and beat types except NSR and AFIB were labeled as OTHER super-class. OTHER contains First-degree atrioventricular blocks, Conduction disturbances, Left bundle branch block, Right bundle branch block, Premature atrial contraction, Premature ventricular contraction, ST-segment depression and ST-segment elevated type ECGs. CurAlive A.I. model which is using DenseNet as a CNN architecture and continuous wavelet transform as feature extraction method, showed a great performance on classifying ECGs from only lead-I compared to cardiologists. The AI model reached the weighted average precision, recall, F1-score and total accuracy 94.1%, 93.6%, 93.7% and 93.6% respectively, and the average of each of the three cardiologists has reached weighted average precision, recall, F1-score and total accuracy 82.2%, 54.6%, 57.5% and 54.6% respectively. This study showed that the proposed CNN model CurAlive, can be used to accurately diagnose AFIB, NSR, and OTHER rhythm using lead-I ECGs to accelerate the early detection of AFIB as a cardiologist assistant. It is also able to identify patients into different risk groups as part of remote patient monitoring systems.

PDF Abstract
No code implementations yet. Submit your code now

Datasets


  Add Datasets introduced or used in this paper

Results from the Paper


  Submit results from this paper to get state-of-the-art GitHub badges and help the community compare results to other papers.

Methods