Abstract:
We present algorithms to distinguish between healthy and diseased condition of the heart, based on the analysis of phonocardiograms. The software tries to mimic the decis...Show MoreMetadata
Abstract:
We present algorithms to distinguish between healthy and diseased condition of the heart, based on the analysis of phonocardiograms. The software tries to mimic the decision-making process of a cardiologist by identifying heart beats (S1 and S2), finding extra sounds and murmurs while ignoring all kinds of artefacts and noise. Two different solutions have been submitted to the PhysioNet Challenge 2016: The entry for phase I aims to reconstruct the signal of an ideal heartbeat by calculating the median of an overlay of all beats of a recording. An LVQ-classifier, trained with the ideal beat of 3240 PCGs of the challenge training set, achieved a specificity of 0.85 and a sensitivity of 0.40, resulting in a total score of 0.63. Our entry for the official phase of the challenge searches for abnormalities in every single beat of a PCG. The results display a sensitivity of 0.91, a specificity of 0.29, and a total score of 0.60.
Published in: 2016 Computing in Cardiology Conference (CinC)
Date of Conference: 11-14 September 2016
Date Added to IEEE Xplore: 02 March 2017
ISBN Information:
Electronic ISSN: 2325-887X
Conference Location: Vancouver, BC, Canada