This work proposes a methodology for identifying heart anomalies on electrocardiograms using a time series novelty detection technique. Novelties or anomalies on time series can be seen as unexpected values or a sequence of unexpected values when compared to a dataset considered to be normal. The technique uses an autoregressive model to estimate the current value of the time series and compares this value with the observed value, if the difference between these two values exceeds one given threshold; the value is classified as a novelty. To build the thresholds this method uses confidence intervals built upon the probability density function of the forecasting model output. In order to treat anomaly as a sequence of data points a post processing technique is applied to the detector output. Experiments were executed on ECG containing anomalies heart beats and the method was able to detect the anomalies with the false alarm expected.
[1]
Gwilym M. Jenkins,et al.
Time series analysis, forecasting and control
,
1971
.
[2]
G. Schwarz.
Estimating the Dimension of a Model
,
1978
.
[3]
Pablo Laguna,et al.
A database for evaluation of algorithms for measurement of QT and other waveform intervals in the ECG
,
1997,
Computers in Cardiology 1997.
[4]
J. Jenkins,et al.
Arrhythmia database for algorithm testing: surface leads plus intracardiac leads for validation.
,
2003,
Journal of electrocardiology.