최대 사후 확률 추정법을 이용한 실시간 생체신호 Peak 검출에 관한 연구

다국어 초록 (Multilingual Abstract)

Real-time Peak Detection for General Biosignals: A Maximum A Posteriori Estimation Approach Yoon Jong Seo Department of Biomedical Engineering Graduate School of Konkuk University With the growing importance of real-time, personalized healthcare through wearable devices, the need for a general-purpose peak detection algorithm that is accurate, efficient, and applicable to various biosignals within resource-constrained environments has become critical. However, existing rule-based and signal processing-based algorithms, while computationally efficient, are often limited to specific signals and lack robustness. Conversely, deep learning-based models, despite their high accuracy, have fundamental limitations of excessive computational load and a lack of generality. To address these issues, this study proposes a novel real-time adaptive peak detection framework based on the 'Maximum a Posteriori (MAP) based Look-Around Algorithm (LAA)', which operates in a single pass without post-processing. The proposed algorithm dynamically adjusts the Look-Around Period (LAP) using the mode of recent RR-intervals to adapt to real-time changes in the cardiac cycle. Furthermore, to handle practical challenges such as T-wave misdetection, peak omission, and R/S waveform ambiguity, the algorithm's robustness is significantly enhanced by incorporating a multi-stage exception handling mechanism based on clear physiological principles, including amplitude comparison, a look-back search, and the 'Preceding Priority' rule. Through rigorous performance verification on the standard benchmark MIT-BIH Arrhythmia Database, the final proposed algorithm achieved an average F1-Score of 98.23%, surpassing the prior study (F1-Score 96.89%) which relied on non-real-time post-processing. Notably, the 'Preceding Priority' rule, based on physiological principles, improved the overall F1-Score by 3.19 percentage points compared to the basic adaptive algorithm and dramatically improved performance on records such as 107, where the empirical rule of the prior study had failed. Moreover, its generality was demonstrated by achieving stable peak detection on PPG and respiration signals with only minimal parameter changes. The significance of this study lies in its proposal of a new design paradigm: a hybrid approach combining an efficient probabilistic model with minimal physiological domain knowledge can be a powerful alternative to computationally intensive deep learning models. This work provides a practical foundation for the development of intelligent on-device healthcare technologies to be embedded in various wearable devices. Keywords: Real-time Peak Detection, General Biosignal Processing, Look-Around Algorithm, Maximum a Posteriori (MAP), Adaptive Algorithm, Wearable Healthcare

목차 (Table of Contents)

• 제1장 서론 1
• 제1절 연구의 배경 및 필요성. 1
• 제2절 기존 연구 및 한계점 2
• 제3절 본 연구의 목적 및 독창성 4
• 제4절 논문의 구성 5
• 제2장 관련 연구. 7
• 제1절 기존 Peak 검출 알고리즘 접근법 7
• 제2절 최대 사후 확률(MAP) 기반 주변 관측 알고리즘 9
• 제3절 연구 공백 및 본 연구의 의의. 12
• 제3장 실시간 주변관측 알고리즘. 13
• 제1절 측정된 신호의 전처리 (Signal Pre-processing). 13
• 제2절 실시간 주변관측 알고리즘. 15
• 1. 알고리즘의 전체 구조. 15
• 2. 주변관측기간의 중요성 17
• 3. 실시간 적응 주변관측 알고리즘 19
• 4. Peak 검출 예외 처리 메커니즘. 21
• 5. 선행우선순위 알고리즘 27
• 제4장 실험 및 결과 30
• 제1절 실험 설계 30
• 1. 실험 데이터베이스 30
• 2. 성능 평가 지표 32
• 제2절 실험 결과 33
• 제5장 고찰 38
• 제1절 결과 분석 및 해석 38
• 제2절 다양한 생체신호 적용 결과 44
• 제3절 향후 연구 방향 46
• 제5장 결론 48
• 참고문헌 50
• 부록 53
• 국문초록 54