Comparative NeuralNetworkBasedVenousThrombosis Echogenicity and Echostructure Characterization UsingUltrasound Images
Chu Cavale Blanche · 2006
Venousthrombosis isa commonpathology that creates serious public health problems. Thrombosis diagnosis, particularly thedetermination oftheir echogenicity andechostructure canbe efficiently accomplished bya medical expert usingultrasound imaging. Ontheother hand, thepredictive capability of artificial neural networks isveryuseful inmedical applications andcansupport medical experts totake appropriate diagnosis decisions. Therefore, the proposed studyintends tocharacterize bymeansof neuralnetworks thethrombosis echogenicity and echostructure, using apredefined learning basethat depends ontheprior knowledge ofphysicians. Wehave studied sixdifferent methodstocharacterize the thrombosis images, along withthesixcorresponding neural networks. Obtained results showthat theoptimal feature vectorsize, thesimplest neuralnetwork architecture, andthesmallest error, areachieved by using themean-variance approach orbythewavelet coefficients energies method.