Fully Automatic Segmentation of Hip CT Images via Random Forest Regression-Based Atlas Selection and Optimal Graph Search-Based Surface Detection

Chu, Chengwen; Bai, Junjie; Liu, Li; Wu, Xiaodong; Zheng, Guoyan (2015). Fully Automatic Segmentation of Hip CT Images via Random Forest Regression-Based Atlas Selection and Optimal Graph Search-Based Surface Detection. In: Cremers, D.; Reid, I.; Saito, H.; Yang, MH. (eds.) 2014 Asian Conference on Computer Vision. Lecture Notes in Computer Science: Vol. 9005 (pp. 640-654). Cham: Springer International Publishing 10.1007/978-3-319-16811-1_42

Text 978-3-319-16811-1_42.pdf - Published Version Restricted to registered users only Available under License Publisher holds Copyright. Download (3MB) | Request a copy

Automatic extraction of surface models of both pelvis and proximal femur of a hip joint from 3D CT images is an important and challenging task for computer assisted diagnosis and planning of periacetabular osteotomy (PAO). Due to the narrowness of hip joint space, the adjacent surfaces of the acetabulum and the femoral head are hardly distinguishable from each other in the target CT images. This paper presents a fully automatic method for segmenting hip CT images using random forest (RF) regression-based atlas selection and optimal graph search-based surface detection. The two fundamental contributions of our method are: (1) An efficient RF regression framework is developed for a fast and accurate landmark detection from the hip CT images. The detected landmarks allow for not only a robust and accurate initialization of the atlases within the target image space but also an effective selection of a subset of atlases for a fast atlas-based segmentation; and (2) 3-D graph theory-based optimal surface detection is used to refine the extraction of the surfaces of the acetabulum and the femoral head with the ultimate goal to preserve hip joint structure and to avoid penetration between the two extracted surfaces. Validation on 30 hip CT images shows that our method achieves high performance in segmenting pelvis, left proximal femur, and right proximal femur with an average accuracy of 0.56 mm, 0.61 mm, and 0.57 mm, respectively.

Interest & Impact

Downloads

0 since deposited on 14 Jul 2017
0 in the past 12 months

Citations

5 Citations in Web of Science ®
6 Citations in Scopus

Search

in Google Scholar™

Services

Actions (login required)

Edit item

Actions (login required)

Edit item