Premium
A Multi‐level Voting Scheme of Kernel Fisher Discriminant Classifiers for Liver Fibrosis Identification in Ultrasound Images
Author(s) -
Salca VictorEugen,
Gordan Mihaela,
Vlaicu Aurel
Publication year - 2011
Publication title -
pamm
Language(s) - English
Resource type - Journals
ISSN - 1617-7061
DOI - 10.1002/pamm.201110418
Subject(s) - artificial intelligence , kernel fisher discriminant analysis , pattern recognition (psychology) , support vector machine , computer science , pixel , ultrasound , kernel (algebra) , feature extraction , linear discriminant analysis , test set , discriminant , classifier (uml) , computer vision , mathematics , medicine , kernel method , radiology , combinatorics
Ultrasound imaging remains important for computer aided medical diagnosis. The assessment of hepatic fibrosis from ultrasound images has been investigated in recent works. Fibrosis modifies the liver texture; this is reflected in the hepatic ultrasound image. The existing works report the extraction of different texture features from small regions of interest (ROIs) in the hepatic ultrasound image used as inputs to machine learning classifiers. As the obtained accuracies are generally moderate (85‐90%), new texture descriptors and classifier schemes are still investigated. Here we assess the performance of the Kernel Fisher Discriminant (KFD) due to its success in other similar tasks. KFD allows an implicit texture discrimination learning from a set of texture samples. The novelty of our solution is its use in a multi‐level scheme. With a training set of ultrasound liver image ROIs of 64×64 pixels, we generate 3 training image sets, one at full size and the others by the decomposition of ROIs into 32×32 and 16×16 pixels blocks. Three KFD classifiers are trained. In the classification phase, each test ROI is classified and the result is obtained by a majority voting scheme. The accuracy of the proposed solution is 89%, on the level of the state of the art, and it can probably be improved by increasing the training set. (© 2011 Wiley‐VCH Verlag GmbH & Co. KGaA, Weinheim)