View : 134 Download: 0

Full metadata record

DC Field Value Language
dc.contributor.advisor유재근-
dc.contributor.author김민정-
dc.creator김민정-
dc.date.accessioned2026-10-01T16:30:06Z-
dc.date.available2026-10-01T16:30:06Z-
dc.date.issued2025-
dc.identifier.otherOAK-000000254284-
dc.identifier.urihttps://dcollection.ewha.ac.kr/common/orgView/000000254284en_US
dc.identifier.urihttps://dspace.ewha.ac.kr/handle/2015.oak/276646-
dc.description.abstractThis dissertation addresses the dual challenges of protecting personal information in large-scale medical data while enhancing analytic reliability. First, we demonstrate how pseudonymization methods, such as data range and aggregation, can preserve essential model performance. Second, a longitudinal study of iron measures in naturally menopausal women revealed that serum ferritin continues to rise well after the final menstrual period, diverging from other iron markers. Third, we investigated FROC-type metrics for evaluating AI-based lesion localization in medical imaging, highlighting the critical roles of detection thresholds and lesion size. Overall, this research provides practical insights for implementing pseudonymization in compliance with privacy regulations, refining clinical interpretation of iron indicators in menopause, and improving AI performance metrics for reliable lesion detection.;의료 및 보건 분야에서 축적되는 대규모 데이터는 개인 정보 보호를 강화해야 한다는 과제와 함께, 통계 · 머신러닝 분석의 신뢰도를 높여야 하는 이중적 요구가 있다. 본 논문은 이를 위해 첫째, 가명처리(pseudonymization) 기법이 자료 유용성을 크게 저해하지 않고 적용될 수 있음을 입증하였다. 둘째, 중년 여성의 폐경 전후 철분 지표 변화를 종단적으로 추적한 결과, 페리틴은 최종 월경 이후에도 장기간 상승을 이어가며 다른 지표와는 상이한 양상을 보이는 점을 발견하였다. 셋째, FROC 계열 지표를 사용해 AI 기반 영상 분석 모델에서 병변 국소화 성능을 평가할 때, 임계값 설정과 병소 크기 등이 결과를 크게 좌우함을 시뮬레이션과 실제 영상 데이터로 확인하였다. 본 연구는 향후 다양한 의료 데이터에서 가명처리를 활용하는 실무적 가이드라인을 제시하고, 폐경기 철분 대사 변화 연구 및 AI 모델 평가 지표 개선에 기여할 것으로 기대한다.-
dc.description.tableofcontentsI. Introduction 1 1. Research Background & Objectives 1 2. Organization of Thesis 3 II. Pseudonymization's Effect on Data Quality: A Study under Personal Information Protection Act 5 1. Summary 5 2. Introduction 5 3. Guidelines for Pseudonymized Data Processing 7 1. Purpose of the Guidelines for Pseudonymized Data Processing 7 2. Terms and Definitions 7 3. Pseudonymization Procedures and Preparatory Measures 9 4. Examples of Pseudonymization Techniques for Personal Information 10 4. Data and Analysis Methods 13 1. Data Description 13 2. Analysis Methods - Logistic Regression, Decision Tree, Random Forest 13 3. Pseudonymization Techniques 14 5. Application of Pseudonymization Techniques 16 1. Application of the Data Range Method 16 2. Application of the Aggregation Method 18 6. Results 21 1. Performance Comparison of Models Using Original Data, Log-Transformed Data, and Pseudonymized Data 21 2. Exploring the Optimal Pseudonymization Ratio 26 7. Discussion 29 III.Longitudinal changes in serum ferritin and other iron-related biomarkers in women during natural menopausal transition 30 1. Summary 30 2. Introduction 30 3. Materials and Methods 31 1. Study population 31 2. Data collection 32 3. Statistical Analysis 34 4. Results 36 1. Participant Characteristics Across Time Segments Linked to the FMP 36 2. Changes in iron measures in relation to the FMP 39 5. Discussion 42 IV. From Pixel Scores to Clinical Outcomes: Revealing the Unseen Decisions in FROC Metric Design 45 1. Summary 45 2. Introduction 45 3. Evaluation Methods 47 1. Evaluation Metrics for Binary Classification 47 2. Evaluation Metrics for Object Detection and Segmentation 48 3. Evaluating Detection Performance 50 4. Design choices for the FROC 51 5. FROC Score Calculation 53 4. Experiments 54 1. Data 54 2. Results 55 3. Interaction between lesion size and thresholds 56 5. Discussion 59 V. Discussion 61 Bibliography 63 Appendix 68-
dc.formatapplication/pdf-
dc.format.extent5673307 bytes-
dc.languageeng-
dc.publisher이화여자대학교 대학원-
dc.subject가명처리| 개인정보 보호| 철 대사 지표| 폐경 전환기| FROC 평가-
dc.subject.ddc500-
dc.titleModern Statistical Challenges in Data Privacy and Complex Healthcare Data-
dc.typeDoctoral Thesis-
dc.creator.othernameKim, Minjeong-
dc.format.pagevi, 68 p.-
dc.identifier.thesisdegreeDoctor-
dc.identifier.major대학원 통계학과-
dc.date.awarded2025. 8-
Appears in Collections:
일반대학원 > 통계학과 > Theses_Ph.D
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

BROWSE