An improved CapsNet applied to recognition of 3D vertebral images
- PDF / 3,597,151 Bytes
- 15 Pages / 595.224 x 790.955 pts Page_size
- 37 Downloads / 197 Views
An improved CapsNet applied to recognition of 3D vertebral images Hao Wang1 · Kun Shao1 · Xing Huo1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Deep learning is currently widely applied in medical image processing and has achieved good results. However, recognizing vertebrae via image processing remains a challenging problem due to their complex spatial structures. CapsNet is a newly proposed network whose characteristics compensate for some shortcomings of traditional CNNs, and it has been shown to perform well on many tasks, including medical image recognition. In this paper, we applied a modified CapsNet to recognise 3D vertebral images by introducing an RNN module into CapsNet to further enhance its learning ability. This new network is called RNNinCaps, and it achieves the highest recognition performance on 3D vertebral images (the average accuracy of RNNinCaps exceeds the accuracy of the original CapsNet by 46.2% and that of a traditional CNN by 12.6%). RNNinCaps also performs better than several mainstream networks. RNNinCaps can promotes CapsNet’s application in the field of 3D medical image recognition. Keywords CapsNet · RNN · CNN · 3D vertebral images · Vertebrae classification
1 Introduction When evaluating spinal health, physicians generally choose imaging techniques such as magnetic resonance (MRI) and computed tomography (CT), because these data provide views of the spinal anatomy [1]. With the development of machine learning technology, deep learning is rapidly becoming a key technique for processing and analysing medical images. At present, studies of tumour recognition [2–6] and pneumonia recognition [7] based on CT images have made significant advances. Automatic identification and localization of vertebral images, which is an important aspect in the field of medical image processing, is also becoming popular because identification and localization are essential factors in the diagnosis and treatment of spinal diseases. Suzani et al. [8] proposed a fast automatic detection and localization method using 2D CT scans; they combined the advantages of a deep neural network for regression and KDE for final centroid estimation to efficiently detect and localize vertebrae in lateral views of the spine. Suzani et al. [9] also proposed Xing Huo
[email protected] 1
Hefei University of Technology, 420 Feicui Road, Hefei, Anhui, China
an automatic method based on deep learning for localising, labelling and segmenting lumbar vertebrae in multislice MR images that achieved good results. However, vertebrae are spatial objects that share similar morphological appearances, and they have complex 3D structures. Although studies such as the two described above are capable of performing some segmentation and recognition tasks, we believe that the spatial structure of 3D data cannot be fully reflected by 2D images; its application scope is also limited (for example, some methods focus only on lumbar vertebrae). It is difficult to distinguish and recognize vertebrae from 2D im
Data Loading...
