Title

Few Scanline Deep Learning Network for Ultrasound Image Segmentation

Alternative Title

적은 수의 스캔라인을 이용한 초음파 이미지 분할 딥러닝 네트워크

DGIST Authors

Minji Kang ; Jae Youn Hwang ; Jin Ho Chang

Advisor

황재윤

Co-Advisor(s)

Jin Ho Chang

Issued Date

2023

Awarded Date

2023-02-01

Citation

Minji Kang. (2023). Few Scanline Deep Learning Network for Ultrasound Image Segmentation. doi: 10.22677/THESIS.200000657134

Type

Thesis

Description

Ultrasound, Wearable System, Segmentation, 초음파, 웨어러블 시스템, 분할

Table Of Contents

1. INTRODUCTION
1.1 Motivation 1
1.2 Background 3
1.2.1 The physics of ultrasound 3
1.2.2 Ultrasound imaging 5
1.2.3 Ultrasound wearable system 6
1.2.4 Deep learning technology 7
1.2.5 Generative Adversarial Network (GAN) 8
1.2.6 Segmentation model 10
1.3 Preliminary Study 13
1.3.1 Correlation between bladder volume and posture 13
1.3.2 Estimation of bladder signal in ultrasound A-mode 15

2. METHODS AND MATERIALS
2.1 Description of the Proposed Bladder Wearable System 17
2.2 Data Acquisition 19
2.2.1 Clinical data acquisition with commercial ultrasound equipment 19
2.2.2 Phantom data acquisition with a single-element transducer 22
2.3 Description of the Bladder Estimation Algorithm 26
2.3.1 Preparation of a paired image dataset 26
2.3.2 Deep learning experiment setup 28
2.3.2.1 GAN (Image-to-image translation) 29
2.3.2.2 Segmentation model (U-Net) 29
2.4 Quantitative Evaluation of Algorithm 30



3. RESULTS AND DISCUSSIONS
3.1 Generated a Larger Number of Scanlines Image 31
3.2 Segmented Bladder Ultrasound Image 34
3.3 Single-element Transducer Experiment Result 41

4. CONCLUSION

REFERENCES 46

요 약 문 49

URI

http://hdl.handle.net/20.500.11750/45745
http://dgist.dcollection.net/common/orgView/200000657134

DOI

10.22677/THESIS.200000657134

Degree

Master

Department

Department of Electrical Engineering and Computer Science

Publisher

DGIST

Show Full Item Record