简介：Osteosarcomaisprimarymalignantneoplasmsderivedfromcellsofmesenchymalorigin,andoftenhasdistinctphenotypesatdifferentstages.Thelocationoftumorandreactionzonecanbeidentifiedbyanexpertinmagneticresonanceimaging(MRI),withMRIbeingoneofthechoicesforevaluatingtheextentofosteosarcoma.However,itisstillachallengetoautomaticallyextracttumorfromitssurroundingtissuesbecauseoftheirlowintensitydifferencesinMRI.WeinvestigatedanapproachbasedonZernikemomentandsupportvectormachine(SVM)forosteosarcomasegmentationinT1-weightedimage(TIWI).Firstly,thedifferentordermomentsaroundeachpixelarecalculatedinsmallwindows.Secondly,thegrayscaleandthemodulevaluesofdifferentordermomentsareusedasatexturefeaturevectorwhichisthenusedasthetrainingsetforSVM.Finally,anSVMclassifieristrainedbasedonthissetoffeaturestoidentifytheosteosarcoma,andthesegmentedtumortissueisrenderedin3Dbytheraycastingalgorithmbasedongraphicsprocessingunit(GPU).TheperformanceofthemethodisvalidatedonT1WI,showingthatthesegmentationmethodhasahighsimilarityindexwiththeexpert’smanualsegmentation.