Title

GAN-based Synthetic Ultrasound Image Generation: Addressing Data Scarcity and Domain Adaptation

Alternative Title

GAN 기반 합성 초음파 이미지 생성: 데이터 부족 및 도메인 적응 문제 해결

DGIST Authors

Eunji Lee ; Jin Ho Chang ; Jae Youn Hwang

Advisor

장진호

Co-Advisor(s)

Jae Youn Hwang

Issued Date

2025

Awarded Date

2025-02-01

Citation

Eunji Lee. (2025). GAN-based Synthetic Ultrasound Image Generation: Addressing Data Scarcity and Domain Adaptation. doi: 10.22677/THESIS.200000837546

Type

Thesis

Description

Deep Learning, Data Augmentation, Domain Adaptation, Ultrasound Image, CycleGAN

Table Of Contents

Ⅰ. Introduction 1
1.1 Background 1
1.2 Related Work 2
1.3 Research Objectives and Approach 7
II. Methods 10
2.1 Network 10
2.1.1 Model Overview 10
2.1.2 Framework 11
2.1.3 Formulation 12
2.2 Datasets 15
2.2.1 Open-Source Datasets 16
2.2.2 Field II Simulation 19
2.3 Training Setup and Details 21
2.3.1 Network Architecture 21
2.3.2 Loss Function 25
2.3.3 Training Setup 27
III. Results 29
3.1 CycleGAN Results 29
3.2 Evaluation 34
3.2.1 Evaluation Metrics 35
3.2.2 Evaluation Method 37
3.2.3 Training Setup 38
3.2.4 Evaluation Results 40
IV. Conclusion 48
4.1 Future Works 48
References 49
국문요약 53

URI

http://hdl.handle.net/20.500.11750/58104
http://dgist.dcollection.net/common/orgView/200000837546

DOI

10.22677/THESIS.200000837546

Degree

Master

Department

Artificial Intelligence Major

Publisher

DGIST

Show Full Item Record