Paper Info
Title
Improvement of tumor detection performance in mammograms by feature selection from a large number of features and proposal of fast feature selection method
Abstract
The purpose of this study is to improve the detection accuracy for malignant tumor shadows on mammograms. More than 490 feature parameters are prepared, more than 10 times the number in previous studies, and the suboptimal feature set which is effective in the detection is selected from among them by the forward stepwise selection procedure. The results are presented in this paper. An experiment using 1698 actual mammograms shows that the number of false-positive malignant tumor shadows is reduced by approximately 40% (from 1.98/image to 1.12/image) compared to the previous. The computational amount of the stepwise selection procedure is theoretically evaluated. It is described that if a larger number of sample images or a larger number of feature parameters than those used in this study are included, the computational amount becomes tremendous, making the process practically impossible. A new fast processing method is proposed, in which the features within the subset are selected in two stages. The new selection procedure is applied to the same data and a feature selection experiment is performed. It is found that the detection accuracy is almost the same and the speed of selection is improved. © 2006 Wiley Periodicals, Inc. Syst Comp Jpn, 37(12): 56–68, 2006; Published online in Wiley InterScience (). DOI 10.1002/scj.20498
Year
DOI
Venue
2006
10.1002/scj.v37:12
Systems and Computers in Japan
Keywords
Field
DocType
feature selection
Stepwise regression,Feature selection,Pattern recognition,Computer science,Feature set,Artificial intelligence,Machine learning
Journal
Volume
Issue
Citations 
37
12
5
PageRank 
References 
Authors
0.51
4
5
Name
Order
Citations
PageRank
Mitsutaka Nemoto1468.42
Akinobu Shimizu250.51
Yoshihiro Hagihara3509.70
Hidefumi Kobatake458653.42
Shigeru Nawano526529.51