eduzhai > Applied Sciences > Engineering >

Fabric Image Representation Encoding Networks for Large-scale 3D Medical Image Analysis

  • Save

... pages left unread,continue reading

Document pages: 12 pages

Abstract: Deep neural networks are parameterised by weights that encode featurerepresentations, whose performance is dictated through generalisation by usinglarge-scale feature-rich datasets. The lack of large-scale labelled 3D medicalimaging datasets restrict constructing such generalised networks. In this work,a novel 3D segmentation network, Fabric Image Representation Networks(FIRENet), is proposed to extract and encode generalisable featurerepresentations from multiple medical image datasets in a large-scale manner.FIRENet learns image specific feature representations by way of 3D fabricnetwork architecture that contains exponential number of sub-architectures tohandle various protocols and coverage of anatomical regions and structures. Thefabric network uses Atrous Spatial Pyramid Pooling (ASPP) extended to 3D toextract local and image-level features at a fine selection of scales. Thefabric is constructed with weighted edges allowing the learnt features todynamically adapt to the training data at an architecture level. Conditionalpadding modules, which are integrated into the network to reinsert voxelsdiscarded by feature pooling, allow the network to inherently processdifferent-size images at their original resolutions. FIRENet was trained forfeature learning via automated semantic segmentation of pelvic structures andobtained a state-of-the-art median DSC score of 0.867. FIRENet was alsosimultaneously trained on MR (Magnatic Resonance) images acquired from 3Dexaminations of musculoskeletal elements in the (hip, knee, shoulder) jointsand a public OAI knee dataset to perform automated segmentation of bone acrossanatomy. Transfer learning was used to show that the features learnt throughthe pelvic segmentation helped achieve improved mean DSC scores of 0.962,0.963, 0.945 and 0.986 for automated segmentation of bone across datasets.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...