Abstract:
—Electrocardiogram (ECG) is a key diagnostic tool to
visualize the heart’s activity and to study its normal or abnormal
functioning. Physicians perform routine diagnosis by visually
examining the shapes of ECG waveform. However, automatic
processing and classification of ECG data would be extremely
useful in patient monitoring and telemedicine systems. Such realtime
applications require techniques that are highly accurate and
very efficient. Most of the literature on ECG data rely on timing
based features for heartbeat classification. This paper presents a
shape or template based method to classify heartbeats as Normal
vs. Premature Ventricular Contraction (PVC) beats which is
capable of being implemented on low computing, low power
consuming and low cost mobile devices such as smartphones. Data
analysis is based on MIT-BIH Arrhythmia Database containing
48 Holter recordings of different patients. An overall accuracy
of 91% was achieved using the proposed method, which is
quite significant considering more than 40,000 heartbeats were
analysed. Furthermore, it was observed that only 3 patients with
peculiar recordings had significantly low accuracies. Excluding
these recordings increased the overall accuracy to 97%. Atypical
nature of these recordings was closely investigated to elicit ideas
for future work
