Meeting Banner
Abstract #3467

Automatic Segmentation of the Medial and Lateral Meniscus in Knee MRI Using Chan-Vese Model with Shape Prior

Junga Baek1, Helen Hong1, Joon Ho Wang2

1Department of Multimedia Engineering, Seoul Women's University, Seoul, Korea; 2Department of Orthopedic Surgery, Samsung Medical Center, Seoul, Korea

For diagnosis of meniscus tears and its reconstruction, we propose an automated segmentation method of the meniscus from knee MR images using Chan-Vese model with shape prior. First, meniscus candidates are extracted by automatically estimating a threshold value using Gaussian Mixture Modeling. Second, cartilage which has similar signal intensity with the meniscus and has a horizontally thin and long shape is removed by shape analysis. Third, Chan-Vese model with shape prior is performed to segment the meniscus without leakage to its neighbor ligament. Our proposed method with shape prior extracts the meniscus without leakage to its neighbor soft tissues such as cartilage and ligament and can be used for the diagnosis of meniscus tears or its reconstruction.

Abstract PDF

Keywords

accuracy adjacent anterior assessed automated automatic automatically candidates cartilage coefficient comparing composed critical developed diagnosis dice difficulties dimensionally distance engineering estimating evaluate evaluation extracted extracts hong horizontally horn intensities intensity knee lateral leakage ligament long manual matrix measured medial medical meniscus mixture model modeling multimedia neighbor original orthopedic outlining overcome performance pixels posterior prior program propose proposed radiologist reconstruction removed resolution scanner segment segmentation segmented segmenting several shape similarity slice slices soft step steps subjects supported surgery symmetric tears thin third threshold tissues variation versus view visually women