가상환자 생성 모델 연구의 동향

• 영문명: Trends in the Studies of Synthetic Patient Records
• 발행기관: 대한미병의학회
• 저자명: 장동엽(Dongyeop Jang) 김창업(Chang-Eop Kim)
• 간행물 정보: 『대한미병의학회지』제1권 제1호, 15~24쪽, 전체 10쪽
• 주제분류: 의약학 > 한의학
• 파일형태: PDF
• 발행일자: 2020.03.01

영문 초록

Objectives With the development of the Internet of things and big data, it becomes possible to analyze large medical records. However, a lack of data and a risk of private information disclosure act as a barrier to this possibility, and generating synthetic patient records can be used as an alternative to circumvent this limitation. In this paper, we reviewed the researches on the rapidly developing synthetic patient records. Methods We searched for articles that studied the synthetic patient records and sorted them according to data types, generation methods and evaluation methods. Results The types of synthetic patient data could be largely divided into binary, ordinal, image, and sequential data. But the studies of models that could handle a dataset of several mixed types were relatively scarce, and more researches are needed. In terms of generation methods, researches could be divided into researches about rule-based models and data- driven model, and the data-driven models could be divided into classical machine learning and deep learning models. The criteria for evaluating the model were largely the reproducibility and risk of privacy disclosure. Both quantitative evaluation and qualitative evaluation were used to evaluate the reproducibility, on the other hand, most the studies lacked privacy assessments and it needs to be supplemented in future studies. Conclusions Applying methods for generating synthetic patient records to Korean traditional medical data will facilitate the data science researches of Korean traditional medicine and this can be an opportunity to understand the structure of Korean traditional medical data.

목차

서론

본론

결론

Conflict of Interest

References

키워드

Synthetic patient records Electronic health records Machine learning Deep learning Big data