We propose a multi-atlas segmentation method for the knee cartilage in T2 PD MR images using sequential multi-atlas registrations and locally-weighted voting (LWV). To select training atlases similar to the test image, a 2D projection image-based atlas selection method is proposed. Then, to extract a bone model to be used as registration target in cartilage segmentation, the bone is segmented by sequential multi-atlas registrations and LWV. Finally, to segment a cartilage without leakage into low-contrast surroundings, the cartilage is segmented by bone-mask-based cartilage registration and shape-constrained LWV with distance and structure similarity weights, as well as atlas similarity weight.
This abstract and the presentation materials are available to members only; a login is required.