디지털 트윈 기반 특이점 회피 경로 계획을 적용한 로봇 초음파 자동 스캐닝을 통한 만성정맥부전 진단 시스템 = Chronic Venous Insufficiency Diagnosis via Automatic Robotic Ultrasound Scanning with Digital Twin-Based Singularity Avoidance Path Planning

다국어 초록 (Multilingual Abstract)

Chronic Venous Insufficiency (CVI) is commonly assessed using duplex ultrasound. However, the diagnostic reliability of this method is hindered by operator’s skill, inconsistent contact pressure, and unstable scanning trajectories. To address these limitations, this study presents a digital twin–based robotic ultrasound system that standardizes image acquisition and enables automatic CVI assessment.
The proposed system identifies the lower-leg region using a stereo camera and pose estimation, ensuring consistent definition of the initial
probe contact point. Point cloud is processed to extract the calf surface and generate scanning waypoints, while surface normal vectors provide orientation angles for stable probe alignment.
A digital twin simulation environment reproduces the manipulator and patient calf geometry, enabling pre-verification of reachability, vision
based collision risk, and singularity-free motion. Validated trajectories are then executed on the real-world manipulator through a Sim-to-Real
approach. During scanning, a hybrid control strategy is used with a constant 20N contact force, and real-time orientation adjustments
improve image uniformity across curved calf geometries.
For vessel extraction, a dual-branch segmentation model processes both raw and enhanced ultrasound images, and attention-based feature
fusion improves boundary clarity. A CNN-based classifier is integrated as a postprocessing filter to ensure anatomically valid deep vein–artery pair detection. CVI diagnosis is performed using the average balance ratio across multiple frames, reducing sensitivity to frame-level noise and enhancing diagnostic stability.
Experimental results demonstrate that the proposed framework improves the reproducibility of ultrasound image acquisition and provides
more consistent CVI evaluation compared to traditional operator dependent methods. The integration of digital twin simulation, robotic
control, and deep learning–based vessel analysis offers a promising foundation for the clinical deployment of autonomous ultrasound systems.

목차 (Table of Contents)

• 1. 서론 1
• 1.1. 연구의 배경 및 목적 1
• 1.2. 기여점 6
• 1.3. 논문의 구성 6
• 2. 만성정맥부전 진단 부위 선정 및 시스템 구성 8
• 2.1. 개요 8
• 2.2. 측정 대상 정맥 선정 근거 8
• 2.3. 진단 시스템 하드웨어 구성 12
• 2.4. 진단 시스템 소프트웨어 구성 15
• 2.5. 디지털 트윈 시뮬레이션 환경 구성 18
• 3. 디지털 트윈 기반 초음파 검사 시스템 20
• 3.1. 개요 20
• 3.2. 다리 인식 알고리즘 20
• 3.2.1. 자세 추정 기반 접촉 지점 정의 22
• 3.2.2. 초음파 스캐닝 경유점 산출 24
• 3.2.3. 스캔 경로 지점 법선 벡터 계산 27
• 3.3. 시뮬레이션 기반 경로 계획 28
• 3.4. 초음파 스캐닝을 위한 매니퓰레이터 제어 알고리즘 34
• 4. 이중분기 세그멘테이션 모델 기반 만성정맥부전 진단 알고리즘 43
• 4.1. 개요 43
• 4.2. 심부정맥-동맥 쌍 세그멘테이션 모델 개발 43
• 4.2.1. 모델 구성 44
• 4.2.2. 준지도 학습 기반 학습 전략 46
• 4.2.3. CNN 기반 분류기를 통한 세그멘테이션 후처리 51
• 4.2.4. 형태학적 분석을 통한 만성정맥부전 진단 52
• 5. 결론 54
• 참고문헌 55