Z-Net: an Anisotropic 3D DCNN for Medical CT Volume Segmentation

Peichao Li,Xiao-Yun Zhou,Zhao-Yang Wang,Guang-Zhong Yang
DOI: https://doi.org/10.1109/iros45743.2020.9341426
2019-01-01
Abstract:Accurate volume segmentation from the Computed Tomography (CT) scan is acommon prerequisite for pre-operative planning, intra-operative guidance andquantitative assessment of therapeutic outcomes in robot-assisted MinimallyInvasive Surgery (MIS). 3D Deep Convolutional Neural Network (DCNN) is a viablesolution for this task, but is memory intensive. Small isotropic patches arecropped from the original and large CT volume to mitigate this issue inpractice, but it may cause discontinuities between the adjacent patches andsevere class-imbalances within individual sub-volumes. This paper presents anew 3D DCNN framework, namely Z-Net, to tackle the discontinuity andclass-imbalance issue by preserving a full field-of-view of the objects in theXY planes using anisotropic spatial separable convolutions. The proposed Z-Netcan be seamlessly integrated into existing 3D DCNNs with isotropic convolutionssuch as 3D U-Net and V-Net, with improved volume segmentation Intersection overUnion (IoU) - up to 12.6%. Detailed validation of Z-Net is provided for CTaortic, liver and lung segmentation, demonstrating the effectiveness andpractical value of Z-Net for intra-operative 3D navigation in robot-assistedMIS.
What problem does this paper attempt to address?