Strict 2-Surface Proximal Classifier with Application to Breast Cancer Detection in Mammograms
Lists
Only you see this
Abstract
We propose a2-plane learning method forbinary classification, namedasthestrict 2-surface proximal (S2SP) classifier, byseeking twocross proximal planes based ontwostrict optimization objectives withasquare ofsumoptimization factor, ofwhich thenonlinearity isachieved byemploying kernel functions. We apply theS2SPclassifier forbothlinear and nonlinear classification torecognize malignant tumors froma setof57regions inmammograms, ofwhich20arerelated to malignant tumors and37tobenign masses. Tendifferent featurecombinations arestudied. Experimental results demonstrate that thelinear S2SPclassifier provides results comparable tothose obtained byFisher linear discriminant analysis (FLDA). Foronefeature set(FS8,seeTable 2),thelinear classification performance wassignificantly improved to0.97 byusing theS2SPclassifier, ascompared totheFLDAperformance of0.82, intermsoftheareaunder thereceiver operating characteristics (ROC)curve. Inthecaseofnonlinear classification, theS2SPclassifier withthetriangle kernel provided aperfect performance of1.0forallofthetenfeature combinations, alsoevaluated intermsoftheareaunder the ROCcurve, butwithgoodrobustness limited tothesetting of thekernel parameter inacertain range. Index Terms-Multiplane learning, proximal classification, square ofsum,breast cancer, breast tumors
BibTeX
@inproceedings{mu2007strict,
author = {Tingting Mu and Asoke K. Nandi and Rangaraj M. Rangayyan},
title = {{Strict 2-Surface Proximal Classifier with Application to Breast Cancer Detection in Mammograms}},
booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
year = {2007},
doi = {10.1109/icassp.2007.366276},
}Loading related content...