Detecting branching structures using local Gaussian models

Wepresentamethodformodellingandestimatingbranchingstructures,such asbloodvesselbifurcations,frommedicalimages.Wemodelbranchesasa superpositionofNGaussianfunctionsinalocalregionwhichdescribethe amplitude,positionandorientationsoflinearfeaturesandalsobranches.The centroidsofcomponentfeaturesareseparatedbyapplyingk-meanstothelocalFourierphaseandthecovariancesandamplitudessubsequentlyestimated byalikelihoodmaximisation.Weapplyapenalisedlikelihoodtest(AIC)to selectthebestfitmodelinaregion.Resultsarepresentedonsyntheticand representative2Dretinalimageswhichshowtheestimationtoberobustand accurateinthepresenceofnoise.Wecompareourresultsascale-spaceoperatormethod.

• Type: Conference
• Year: 2002
• Where: ISBI
• Authors: Li Wang, Abhir Bhalerao
• Keywords: Frommedicalimages.Wemodelbranchesasa SuperpositionofNGaussianfunctions, ISBI 2002, Medical Imaging, Such Asbloodvesselbifurcations, Wepresentamethodformodellingandestimatingbranchingstructures