Detection andClassification ofCardiac Murmurs using Segmentation Techniques andArtificial Neural Networks

Inthis paperwepresent theimplementation ofa diagnostic systembasedon Artificial NeuralNetworks (ANN)that canbeusedinthedetection andclassification of heartmurmurs.Segmentation andalignment algorithms serveasimportant pre-processing steps before heart sounds areapplied totheANN structure. Thesystem enables users tocreate aclassifier thatcanbetrained todetect virtually anydesired target setofheartsounds. Theoutput ofthe system istheclassification ofthesoundaseither normalor atypeofheart murmur.Theultimate goalofthis research is toimplement aheart soundsdiagnostic system thatcanbe usedtohelp physicians intheauscultation ofpatients andto reduce thenumberofunnecessary echocardiograms-- those thatareordered forhealthy patients. Testing hasbeen conducted using bothsimulated andrecorded patient heart soundsasinput. Threesetsofresults forthetested system areincluded herein, corresponding tothree different target setsofsimulated heart sounds. Thesystem isabletoclassify withupto85± 7.40% accuracy and95±6.80% sensitivity. Foreachtarget set, theaccuracy rateoftheANN system is compared totheaccuracy rateofagroupof2nd yearmedical students whowereaskedtoclassify heart sounds fromthe samegroupofheart soundsclassified bytheANN system. System test results arealsoexplored using recorded patient heart sounds.