eduzhai > Applied Sciences > Engineering >

Fully Automated 3D Segmentation of MR-Imaged Calf Muscle Compartments Neighborhood Relationship Enhanced Fully Convolutional Network

  • Save

... pages left unread,continue reading

Document pages: 28 pages

Abstract: Automated segmentation of individual calf muscle compartments from 3Dmagnetic resonance (MR) images is essential for developing quantitativebiomarkers for muscular disease progression and its prediction. Achievingclinically acceptable results is a challenging task due to large variations inmuscle shape and MR appearance. Although deep convolutional neural networks(DCNNs) achieved improved accuracy in various image segmentation tasks, certainproblems such as utilizing long-range information and incorporating high-levelconstraints remain unsolved. We present a novel fully convolutional network(FCN), called FilterNet, that utilizes contextual information in a largeneighborhood and embeds edge-aware constraints for individual calf musclecompartment segmentations. An encoder-decoder architecture with flexiblebackbone blocks is used to systematically enlarge convolution receptive fieldand preserve information at all resolutions. Edge positions derived from theFCN output muscle probability maps are explicitly regularized usingkernel-based edge detection in an end-to-end optimization framework. OurFilterNet was evaluated on 40 T1-weighted MR images of 10 healthy and 30diseased subjects by 4-fold cross-validation. Mean DICE coefficients of88.00 --91.29 and mean absolute surface positioning errors of 1.04--1.66 mmwere achieved for the five 3D muscle compartments.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×