| | |
Summary: Abstract-- Musculoskeletal applications of MRI are
increasing rapidly but a major challenge for researchers is
the ability to efficiently and accurately segment structures of
interest, such as bone, which is typically required to perform
further quantitative analyses. Manual tracing is extremely
time consuming and introduces problematic user variability.
Automated segmentation is usually preferred; however, the
accuracy and robustness of current methods still suffer from
significant limitations. In this paper, we propose a novel
approach for simplifying such segmentation tasks by
optimizing MR protocols specifically for bone data
acquisition. We present multi-contrast MR bone data
acquired using short-TR T1W and fat suppression scans and
demonstrate how this data can be used within an automated
segmentation framework in order to improve accuracy of
bone segmentation. Validation was performed on knee joint
data with quantitative segmentation results on our multi-
contrast data showing superior performance compared to
results obtained using conventional single-contrast data.
Improvements in contrast to noise ratio of 39.24 and in
|
