Meeting Banner
Abstract #2561

Method for Improving Segmentation of Multispectral Brain MRI by a Supervised Hybrid Classifier

Jyh-Wen Chai1, Clayton Chi-Chang Chen1, Hsian-Min Chen2, Yaw-Jiunn Chiou3, Shih-Yu Chen4, Yi-Ying Wu1, Chih-Ming Chiang1, Ching-Wen Yang5, Yen-Chieh Ouyang6, San-Kan Lee1, Chein_I Chang4

1Department of Radiology, Taichung Veterans General Hospital, Taichung, Taiwan; 2Department of Biomedical Engineering, HungKung university, Taichung, Taiwan; 3Department of Electrical Engineering, National Chung Hsing University; 4Department of Computer Science and Electrical Engineering, University of Maryland, Baltimore County, Baltimore, MD; 5Computer Center, Taichung Veterans General Hospital, Taichung, Taiwan; 6Department of Electrical Engineering, National Chung Hsing University, Taichung, Taiwan

A hybrid classifier, derived from iterative Fishers linear discriminant analysis coupled with the volume sphering analysis and support vecter machine, was developed to effectively segment multi-slice data of multispectral brain MRI by using only one set of training samples. The proposed method has several advantages. One was a reduction of computational cost in data processing since it only needs one set of training samples to process the entire multislice images. Besides, the same saving is also applied in minimizing operator burden. The uppermost benefit is to avoid operator interferences from selecting training samples and improve the reproducibility.

Abstract PDF

Keywords

accuracy advantages analyzed applicability applicable applications applied avoid background besides biomedical brain burden classification classified classifier classify clinical coefficients commonly component components computational computer conduct consecutive consistency consists containing cost county coupled crucial date density depicted derived discriminant effective effectively electrical engineering enhancement entire evaluate experienced forth general hart healthy hospital hybrid identical implement improve improvement improving independent index indexes individual intensity inter interferences intra iterations iterative john lack last lesions limitations linear listed matrix meaningful modified much national needed needs next noise operator operators paper pathologies potential practicality preprocessing previous process processes processing produced proposed proton quantification quantitative radiologist radiology real reduction removing reproducibility sample samples saving scene segment segmentation selected selecting selection sensitivity served sets settings several simulated simultaneously since slice sons sphered sphering stage statistics step structures substances superior supervised support synthetic table third tissue tissues training twenty uppermost variability variation various vector veterans volume volumetric volunteers yang