2.75D: Boosting learning by representing 3D Medical imaging to 2D features for small data

Wang，Xin1; Su，Ruisheng2; Xie，Weiyi3; Wang，Wenjin4; Xu，Yi5; Mann，Ritse1; Han，Jungong6; Tan，Tao1,7

2023-07-01

10.1016/j.bspc.2023.104858

Biomedical Signal Processing and Control   影响因子和分区

1746-8094

1746-8108

In medical-data driven learning, 3D convolutional neural networks (CNNs) have started to show superior performance to 2D CNNs in numerous deep learning tasks, proving the added value of 3D spatial information in feature representation. However, the difficulty in collecting more training samples to converge, more computational resources and longer execution time make this approach less applied. Also, applying transfer learning on 3D CNN is challenging due to a lack of publicly available pre-trained 3D models. To tackle these issues, we proposed a novel 2D strategical representation of volumetric data, namely 2.75D. In this work, the spatial information of 3D images is captured in a single 2D view by a spiral-spinning technique. As a result, 2D CNN networks can also be used to learn volumetric information. Besides, we can fully leverage pre-trained 2D CNNs for downstream vision problems. We also explore a multi-view 2.75D strategy, 2.75D 3 channels (2.75D × 3), to boost the advantage of 2.75D. We evaluated the proposed methods on three public datasets with different modalities or organs (Lung CT, Breast MRI, and Prostate MRI), against their 2D, 2.5D, and 3D counterparts in classification tasks. Results show that the proposed methods significantly outperform other counterparts when all methods were trained from scratch on the lung dataset. Such performance gain is more pronounced with transfer learning or in the case of limited training data. Our methods also achieved comparable performance on other datasets. In addition, our methods achieved a substantial reduction in time consumption of training and inference compared with the 2.5D or 3D method.

2.75D Breast cancer CT Deep learning Luna cancer Medical imaging MRI Prostate cancer Spiral sampling

SCI ; EI

英语

其他

Macao Polytechnic University Grant[RP/FCA-15/2022]

Engineering

Engineering, Biomedical

WOS:000972663800001

ELSEVIER SCI LTD

20231213752722

Biological organs ; Classification (of information) ; Computerized tomography ; Convolutional neural networks ; Deep learning ; Diseases ; Magnetic resonance imaging ; Urology ; Volumetric analysis

Biomedical Engineering:461.1 ; Biological Materials and Tissue Engineering:461.2 ; Ergonomics and Human Factors Engineering:461.4 ; Medicine and Pharmacology:461.6 ; Magnetism: Basic Concepts and Phenomena:701.2 ; Information Theory and Signal Processing:716.1 ; Computer Applications:723.5 ; Imaging Techniques:746 ; Chemistry:801 ; Information Sources and Analysis:903.1

2-s2.0-85150294572

Scopus

1.Department of Radiology,The Netherlands Cancer Institute,Amsterdam,Plesmanlaan 121,1066 CX,Netherlands
2.Erasmus Medical Center Rotterdam,Rotterdam,Doctor Molewaterplein 40,3015 CD,Netherlands
3.Radboud University Medical Center,Nijmegen,Geert Grooteplein Zuid 10,6525 GA,Netherlands
4.Biomedical Engineering Department of Southern University of Science and Technology,Shenzhen,Xueyuan Blvd 1088,518055,China
5.Shanghai Key Lab of Digital Media Processing and Transmission,Shanghai Jiao Tong University,Shanghai,Dong Chuan Road 800,200240,China
6.Department of Computer Science,the University of Sheffield,Sheffield,Western Bank,S10 2TN,United Kingdom
7.Faculty of Applied Sciences,Macao Polytechnic University,Rua de Luís Gonzaga Gomes,China

Wang，Xin,Su，Ruisheng,Xie，Weiyi,et al. 2.75D: Boosting learning by representing 3D Medical imaging to 2D features for small data[J]. Biomedical Signal Processing and Control,2023,84.

Wang，Xin.,Su，Ruisheng.,Xie，Weiyi.,Wang，Wenjin.,Xu，Yi.,...&Tan，Tao.(2023).2.75D: Boosting learning by representing 3D Medical imaging to 2D features for small data.Biomedical Signal Processing and Control,84.

Wang，Xin,et al."2.75D: Boosting learning by representing 3D Medical imaging to 2D features for small data".Biomedical Signal Processing and Control 84(2023).