TY - GEN
T1 - Diagnosis of AF based on time and frequency features by using a hierarchical classifier
AU - Liu, Yang
AU - Wang, Kuanquan
AU - Li, Qince
AU - He, Runnan
AU - Xia, Yong
AU - Li, Zhen
AU - Liu, Hao
AU - Zhang, Henggui
N1 - Publisher Copyright:
© 2017 IEEE Computer Society. All rights reserved.
PY - 2017
Y1 - 2017
N2 - Early diagnosis of Atrial Fibrillation (AF) could be benefited from automatic analysis of a short single-lead ECG recording that can be collected easily by a portable device. Due to the limitations of both quantity and quality of the signal, it is challenging to distinguish AF from a broad taxonomy of rhythms. This paper presents a new method which classifies the recordings of single lead ECGs by combined time and time-frequency features. The time features of a recording are represented by some characteristics of its RR intervals and Poincare plot, while the time-frequency features are extracted from its representative beat waveforms by Matching Pursuits algorithm. A set of methods are adopted in the process to eliminate the effects of noise. With the features extracted, a hierarchical classifier is trained based on the CinC Challenge 2017 dataset to classify the recordings into four classes: normal sinus rhythm, AF, other rhythm and too noisy to classify. The final score of our work in the CinC Challenge 2017 is 0.78.
AB - Early diagnosis of Atrial Fibrillation (AF) could be benefited from automatic analysis of a short single-lead ECG recording that can be collected easily by a portable device. Due to the limitations of both quantity and quality of the signal, it is challenging to distinguish AF from a broad taxonomy of rhythms. This paper presents a new method which classifies the recordings of single lead ECGs by combined time and time-frequency features. The time features of a recording are represented by some characteristics of its RR intervals and Poincare plot, while the time-frequency features are extracted from its representative beat waveforms by Matching Pursuits algorithm. A set of methods are adopted in the process to eliminate the effects of noise. With the features extracted, a hierarchical classifier is trained based on the CinC Challenge 2017 dataset to classify the recordings into four classes: normal sinus rhythm, AF, other rhythm and too noisy to classify. The final score of our work in the CinC Challenge 2017 is 0.78.
UR - https://www.scopus.com/pages/publications/85045085959
U2 - 10.22489/CinC.2017.180-102
DO - 10.22489/CinC.2017.180-102
M3 - 会议稿件
AN - SCOPUS:85045085959
SN - 9781538645550
T3 - Computing in Cardiology
SP - 1
EP - 4
BT - Computing in Cardiology 2017, CinC 2017
PB - IEEE Computer Society
T2 - 44th Computing in Cardiology Conference, CinC 2017
Y2 - 24 September 2017 through 27 September 2017
ER -