期刊
MEDICAL ENGINEERING & PHYSICS
卷 27, 期 5, 页码 415-424出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.medengphy.2004.10.003
关键词
pattern recognition; segmentation; discriminant analysis; Legg-Calve-Perthes disease; hip joint
This study proposes semi-automatic determination of geometrical features in hip magnetic resonance (MR) images in order to evaluate the Legg Calve-Perthes disease (LCPD). Nine anatomical points on a hip image are selected by a clinician; then eight geometrical indexes of the hip joint are calculated-acetabulium head index (AHI), Wiberg angle (VCE), inner acetabular coverage angle (VCI), acetabular inclination angle (HTE), femoral shaft-neck angle (CC'D), circularity (C), convex deficiency factor (CDF) and pillar height deficiency factor (HDF) for the head region. The geometrical parameters are evaluated on 46 hip images of young patients with unilateral LCPD: 23 images concern the affected hip and 23 the unaffected hip. The extraction of the region of interest is done with a seeded region growing method. All the data were centered and reduced, and were subjected to principal component analysis. Supervised classification is applied with discriminant analysis and k-nearest neighbours classification. The AHI appears to be the best discriminant attribute (maximum between-class variance ratio). Cross-validation tests indicate that we can at most reduce the parameters to five (AHI, CC'D, DHF, DCF and VCE). The classification error rate for the linear discriminant method is 12.5%. (c) 2004 IPEM. Published by Elsevier Ltd. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据