A Deformable Model using Probabilistic Labelling and Surface Relaxation to Segment MR Volumes

Naomi Hill , Roger Boyle and Elizabeth Berry
School of Computer Studies, University of Leeds
Centre of Medical Imaging Research, University of Leeds
naomi@scs.leeds.ac.uk

Abstract:

Deformable models often require some degree of user-interaction to produce an accurate segmentation of the feature object. We propose an automatic deformable surface for segmenting the femur from magnetic resonance volumes of the knee.

In order to overcome the limitations of many existing deformable models, probabilistic labelling is used to drive the deformation process, together with a surface relaxation phase to prevent self-intersection and leakage. Results are presented for the segmentation of the femur from knee MR volumes.

 
• 1 Introduction
• 2 Knee MR data
• 3 The Deformable Surface
• 4 Results
• 5 Conclusion
• References