文献类型： 会议论文

第一作者： Karla Horsch

作者： Karla Horsch ¹ ; Alfredo Fredy Ceballos ¹ ; Maryellen L. Giger ¹ ; Ioana Bonta ¹ ; Zhimin Huo ¹ ; Carl J. Vyborny ¹ ; Edward Hendrick ² ; Li Lan ¹ ;

作者机构： 1.Department of Radiology, University of Chicago, Chicago, IL

2.Lynn Sage Breast Center, Northwestern University, Chicago, IL

关键词： feature selection;computer-aided diagnosis;mammography;ultrasound

会议名称： Image processing :

主办单位：

页码： 986-993

摘要： Linear step-wise feature selection is performed for computerized analysis methods on a set of mammography features using a database of mammography cases, a set of ultrasound features using a database of ultrasound cases, and a set of mammography and sonography features using a multi-modality database were randomly split 20 times into three subdatabases for feature selection, classifier training and independent validation. The average validation A_z value over the 20 random splits for the mammography database was 0.82 ± 0.04 and for the sonography database was 0.85 ± 0.03. The average consistency feature selection A_z value for the mammograhy and sonography databases were 0.87 ± 0.02 and 0.88 ± 0.02, respectively. For the multi-modality database, the consistency feature selection A_z value was 0.93.

分类号： N55