USCL: Pretraining Deep Ultrasound Image Diagnosis Model Through Video Contrastive Representation Learning

Chen，Yixiong1,4; Zhang，Chunhui2; Liu，Li3,4; Feng，Cheng5,6; Dong，Changfeng5,6; Luo，Yongfang5,6; Wan，Xiang3,4

10.1007/978-3-030-87237-3_60

2021

International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)

0302-9743

1611-3349

978-3-030-87236-6

LECT NOTES ARTIF INT

12908

627-637

SEP 27-OCT 01, 2021

null,null,ELECTR NETWORK

GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND

SPRINGER INTERNATIONAL PUBLISHING AG

Most deep neural networks (DNNs) based ultrasound (US) medical image analysis models use pretrained backbones (e.g., ImageNet) for better model generalization. However, the domain gap between natural and medical images causes an inevitable performance bottleneck. To alleviate this problem, an US dataset named US-4 is constructed for direct pretraining on the same domain. It contains over 23,000 images from four US video sub-datasets. To learn robust features from US-4, we propose an US semi-supervised contrastive learning method, named USCL, for pretraining. In order to avoid high similarities between negative pairs as well as mine abundant visual features from limited US videos, USCL adopts a sample pair generation method to enrich the feature involved in a single step of contrastive optimization. Extensive experiments on several downstream tasks show the superiority of USCL pretraining against ImageNet pretraining and other state-of-the-art (SOTA) pretraining approaches. In particular, USCL pretrained backbone achieves fine-tuning accuracy of over 94% on POCUS dataset, which is 10% higher than 84% of the ImageNet pretrained model. The source codes of this work are available at https://github.com/983632847/USCL.

Contrastive learning Pretrained model Ultrasound

其他

英语

CPCI-S ; EI

Key-Area Research and Development Program of Guangdong Province[2020B0101350001] ; GuangDong Basic and Applied Basic Research Foundation[2020A1515110376]

Acoustics ; Computer Science ; Engineering ; General & Internal Medicine ; Microscopy ; Radiology, Nuclear Medicine & Medical Imaging

Acoustics ; Computer Science, Artificial Intelligence ; Engineering, Biomedical ; Medicine, General & Internal ; Microscopy ; Radiology, Nuclear Medicine & Medical Imaging

WOS:000712019200060

20214110994527

Deep neural networks ; Diagnosis ; Medical imaging

Biomedical Engineering:461.1 ; Ergonomics and Human Factors Engineering:461.4 ; Medicine and Pharmacology:461.6 ; Imaging Techniques:746 ; Ultrasonic Waves:753.1

2-s2.0-85116489624

Scopus

1.School of Data Science,Fudan university,Shanghai,China
2.Institute of Information Engineering,Chinese Academy of Sciences,Beijing,China
3.Shenzhen Research Institute of Big Data,Shenzhen,China
4.The Chinese University of Hong Kong Shenzhen,Shenzhen,China
5.Shenzhen Third People’s Hospital,Shenzhen,China
6.Southern University of Science and Technology,Shenzhen,China

Chen，Yixiong,Zhang，Chunhui,Liu，Li,et al. USCL: Pretraining Deep Ultrasound Image Diagnosis Model Through Video Contrastive Representation Learning[C]. GEWERBESTRASSE 11, CHAM, CH-6330, SWITZERLAND:SPRINGER INTERNATIONAL PUBLISHING AG,2021:627-637.