CMStatistics 2022: Start Registration
View Submission - CMStatistics
Title: Considering errors in ECG segmentation to improve racehorse arrhythmia detection Authors:  Camille Meneur - ENSAI (France) [presenting]
Guillaume Dubois - LIM group (France)
Sandrine Hanne-Poujade - LIM group (France)
Matthieu Marbac - CREST - ENSAI (France)
Pauline Martin - LIM group (France)
Gilles Stupfler - University of Angers (France)
Abstract: Electrocardiograms (ECG) are signals that are composed of consecutive beat sequences. Arrhythmias are investigated with ECG and defined by an excessive length of a beat sequence. The real segmentation of the ECG being not observed, segmentation algorithms must be used but can produce errors in the segmentation that deteriorate the detection of arrhythmia. The aim is to take into account the errors made by the segmentation algorithm in order to improve the detection of arrhythmia. This problem can be interpreted as a problem of homogeneity testing made on the distribution of the length of the beat sequences, when this variable suffers from measurement errors. Indeed, the distribution of the length of a beat sequence depends on a latent variable that indicates whether the beat is arrhythmic. To circumvent the issue of measurement errors, we proposed to add, in the homogeneity test, other features extracted from the original signal. Results show that the proposed method permits to differentiate errors from arrhythmic and healthy sequences, and thus to reduce the number of type $I$ and type $II$ errors. This method is tested on racehorses ECG recorded during training.