OnlyCaps-Net, a Capsule only Based Neural Network for 2D and 3D Semantic Segmentation

被引:1
作者
Bonheur, Savinien [1 ,2 ]
Thaler, Franz [2 ,4 ]
Pienn, Michael [1 ]
Olschewski, Horst [1 ,3 ]
Bischof, Horst [2 ]
Urschler, Martin [5 ]
机构
[1] Ludwig Boltzmann Inst Lung Vasc Res, Graz, Austria
[2] Graz Univ Technol, Inst Comp Graph & Vis, Graz, Austria
[3] Med Univ Graz, Dept Internal Med, Graz, Austria
[4] Med Univ Graz, Gottfried Schatz Res Ctr Biophys, Graz, Austria
[5] Univ Auckland, Sch Comp Sci, Auckland, New Zealand
来源
MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION, MICCAI 2022, PT V | 2022年 / 13435卷
关键词
Capsule networks; Convolutional Neural Networks; Multi-label; Semantic segmentation; 2D; 3D; CHEST RADIOGRAPHS;
D O I
10.1007/978-3-031-16443-9_33
中图分类号
TP39 [计算机的应用];
学科分类号
081203 ; 0835 ;
摘要
Since their introduction by Sabour et al., capsule networks have been extended to 2D semantic segmentation with the introduction of convolutional capsules. While extended further to 3D semantic segmentation when mixed with Convolutional Neural Networks (CNNs), no capsule-only network (to the best of our knowledge) has been able to reach CNNs' accuracy on multilabel segmentation tasks. In this work, we propose OnlyCaps-Net, the first competitive capsule-only network for 2D and 3D multi-label semantic segmentation. OnlyCaps-Net improves both capsules' accuracy and inference speed by replacing Sabour et al. squashing with the introduction of two novel squashing functions, i.e. softsquash or unitsquash, and the iterative routing with a new parameter free single pass routing, i.e. unit routing. Additionally, OnlyCapsNet introduces a new parameter efficient convolutional capsule type, i.e. depthwise separable convolutional capsule.
引用
收藏
页码:340 / 349
页数:10
相关论文
共 12 条
[1]  
[Anonymous], ADADELTA: An Adaptive Learning Rate Method
[2]  
Bonheur S., 2019, INT C MEDICAL IMAGE
[3]  
Chen TQ, 2016, Arxiv, DOI [arXiv:1604.06174, DOI 10.48550/ARXIV.1604.06174]
[4]  
Simpson AL, 2019, Arxiv, DOI arXiv:1902.09063
[5]  
LaLonde R., 2018, Capsules for object segmentation
[6]  
Padoy N., 2021, MEDICAL IMAGE COMPUT
[7]   Multi-label Whole Heart Segmentation Using CNNs and Anatomical Label Configurations [J].
Payer, Christian ;
Stern, Darko ;
Bischof, Horst ;
Urschler, Martin .
STATISTICAL ATLASES AND COMPUTATIONAL MODELS OF THE HEART: ACDC AND MMWHS CHALLENGES, 2018, 10663 :190-198
[8]  
Sabour S, 2017, ADV NEUR IN, V30
[9]   Development of a digital image database for chest radiographs with and without a lung nodule: Receiver operating characteristic analysis of radiologists' detection of pulmonary nodules [J].
Shiraishi, J ;
Katsuragawa, S ;
Ikezoe, J ;
Matsumoto, T ;
Kobayashi, T ;
Komatsu, K ;
Matsui, M ;
Fujita, H ;
Kodera, Y ;
Doi, K .
AMERICAN JOURNAL OF ROENTGENOLOGY, 2000, 174 (01) :71-74
[10]   Segmentation of anatomical structures in chest radiographs using supervised methods: a comparative study on a public database [J].
van Ginneken, B ;
Stegmann, MB ;
Loog, M .
MEDICAL IMAGE ANALYSIS, 2006, 10 (01) :19-40