Analysis of echocardiographic image sequences to study left ventricular performance

Abstract

Cardiovasculardiseasesarepathologiesthataffecttheheartandbloodvessels.According to theworldhealthorganization,theyaretheleadingcauseofmortalityworldwide. Early diagnosisofcardiacfunctiondisordersiscrucialinreducingthemortalityrate. The LeftVentricle(LV)isavitalcomponentofthecardiovascularsystemandplaysa significantroleinbloodcirculation.Severalclinicalparameterscanbeestimatedfrom the LVstructureduringcardiovascularexamstoensurereliablediagnoses,includingleft ventricularvolumesandejectionfraction. Variouscardiacimagingmodalitiesallowvisualizationoftheleftventricularcavity. Echocardiographyisthemostwidelyusedtechniquebycardiologistsinroutineclinical practice duetoitsmanyadvantages.Theprimarymethodforestimatingclinicalpa- rameters isLVsurfacesegmentationfrom2Dechocardiographicimagesequences.The accurate evaluationoftheLVchamber’sfunctionreliesonthequalityofthesegmentation results. However,LVmanualdelineationbycardiologistsisdifficult,time-consuming,and imprecise duetothelowqualityofechocardiographicimages.Therefore,thereisaneed to automaticallysegmenttheLVfromechocardiographicimagesequencestoovercome these challenges. In thisthesis,ourobjectiveistodevelopafullyautomaticsegmentationframework based ondeeplearningtechniquestoassessLVperformanceusingechocardiographicim- ages. Wetestedtheeffectivenessoftheproposedapproachesbycomparingtheobtained results withgroundtruthdataandexistingstate-of-the-artmethodsinthisfield.The results aresatisfactory,underliningthesignificantpotentialofautomatedtechniquesfor echocardiographicimageanalysistohelpcardiologistsintheirdailyclinicalpractice.