%0 Journal Article
%T Auto-Segmentation on Liver with U-Net and Pixel De-Convolutional Network
%A Huan Yao
%A Jenghwa Chang
%J International Journal of Medical Physics,Clinical Engineering and Radiation Oncology
%P 81-93
%@ 2168-5444
%D 2021
%I Scientific Research Publishing
%R 10.4236/ijmpcero.2021.102008
%X <strong>Purpose</strong><span style=\"font-family:;\" \"=\"\"><span style=\"font-family:Verdana;\"><strong>:&nbsp;</strong></span><span style=\"font-family:Verdana;\">To improve the liver
auto-segmentation performance of three-</span><span style=\"font-family:Verdana;\">dimensional (3D) U-net by replacing
the conventional up-sampling convolution layers with the Pixel De-convolutional
Network (PDN) that considers spatial features. </span><b><span style=\"font-family:Verdana;\">Methods</span></b><span style=\"font-family:Verdana;\">: The U-net was
originally developed to segment neuronal structure with outstanding performance
but suffered serious artifacts from indirectly unrelated adjacent pixels in its
up-sampling layers. The hypothesis of this study was that the segmentation
quality of </span></span><span style=\"font-family:Verdana;\">the </span><span style=\"font-family:Verdana;\">liver could be
improved with PDN in which the up-sampling layer was replaced by a pixel
de-convolution layer (PDL). Seventy</span><span style=\"font-family:Verdana;\">-</span><span style=\"font-family:;\" \"=\"\"><span style=\"font-family:Verdana;\">eight plans of abdominal cancer patients were
anonymized and exported. Sixty-two were chosen for training two networks: 1) 3D
U-Net, and 2) 3D PDN, by minimizing the Dice loss function. The other sixteen
plans were used to test the performance. The similarity Dice and Average
Hausdorff Distance (AHD) were calculated and compared between these two
networks. </span><b><span style=\"font-family:Verdana;\">Results</span></b><span style=\"font-family:Verdana;\">: The computation time for 62 training cases and 200
training epochs was about 30 minutes for both networks. The segmentation
performance was evaluated using the remaining 16 cases. For the Dice score, the
mean ¡À standard deviation were 0.857 ¡À 0.011 and 0.858 ¡À 0.015 for the PDN and
U-Net, respectively. For the AHD, the mean ¡À standard deviation were 1.575 ¡À
0.373 and 1.675 ¡À 0.769, respectively, corresponding to an improvement of 6.0%
and 51.5% of mean and standard deviation for the PDN. </span><b><span style=\"font-family:Verdana;\">Conclusion</span></b><span style=\"font-family:Verdana;\">: The
PDN has outperformed the U-Net on liver auto-segmentation. The predicted
contours of PDN are more conformal and smoother when compared with</span></span><span style=\"font-family:Verdana;\"> the</span><span style=\"font-family:Verdana;\"> U-Net.</span>
%K Liver Auto-Segmentation
%K Deep-Learning
%K U-Net
%K Pixel-Deconvolutional Network
%U http://www.scirp.org/journal/PaperInformation.aspx?PaperID=109449