A Self-supervised Deep Neural Network for Seamless Image Reconstruction of a Robot- assisted Ultrasound System
로봇 보조 초음파 시스템의 영상 재구성을 위한 자가지도 학습 심층 신경망
- 주제(키워드) self-supervised , neural network , robot-assisted ultrasound system , image registration
- 발행기관 서강대학교 일반대학원
- 지도교수 유양모
- 발행년도 2026
- 학위수여년월 2026. 2
- 학위명 석사
- 학과 및 전공 일반대학원 전자공학과
- 실제URI http://www.dcollection.net/handler/sogang/000000082689
- UCI I804:11029-000000082689
- 본문언어 영어
- 저작권 논문은 저작권에 의해 보호받습니다.
초록(요약문)
Robot-Assisted Ultrasound (RAUS) systems have been developed to overcome the operator dependency and low reproducibility of handheld ultrasound by leveraging precise robotic control and spatial data. Despite these advantages, high-fidelity 3D reconstruction remains challenging because robotic coordinate inaccuracies and non-rigid tissue deformation cause significant misalignments between captured frames. To overcome these limitations, this study proposes a RAUS system integrated with a self-supervised deep neural network (DNN) for seamless, wide-view image reconstruction. By leveraging a self-supervised learning framework, the neural network is trained effectively without labeled datasets and successfully compensates for misalignments induced by robotic coordinate inaccuracies and non-rigid tissue deformation. Experimental results demonstrate that the neural network achieves superior registration accuracy, outperforming rigid and non-rigid image registration methods in both qualitative and quantitative assessments. Furthermore, the integrated RAUS system, empowered by this neural network, successfully eliminates artifacts and voids, producing high-quality volumetric images. This research presents a robust, automated solution that enhances the clinical utility of RAUS by enabling accurate, real-time non-rigid registration.
more목차
Abstract 5
I. Introduction 6
A. Robot-assisted ultrasound 6
B. Deep learning approaches in ultrasound image registration 8
C. Research motivation 10
II. Methods and Material 12
A. Self-supervised learning framework 12
B. Proposed robot-assisted ultrasound system 20
III. Experimental setups 29
A. Dataset utilization 29
B. Training Parameters 30
C. Evaluation metrics 31
IV. Experimental result 33
A. Image registration results 33
B. 3D reconstruction results 35
V. Discussion and Conclusions 37
A. Discussion 37
B. Conclusions 38
Reference 39
