Review learning: Real world validation of privacy preserving continual learning across medical institutions.

Journal: Computers in biology and medicine
Published Date:

Abstract

When a deep learning model is trained sequentially on different datasets, it often forgets the knowledge learned from previous data, a problem known as catastrophic forgetting. This damages the model's performance on diverse datasets, which is critical in privacy-preserving deep learning (PPDL) applications based on transfer learning (TL). To overcome this, we introduce "review learning" (RevL), a low cost continual learning algorithm for diagnosis prediction using electronic health records (EHR) within a PPDL framework. RevL generates data samples from the model which are used to review knowledge from previous datasets. Six simulated institutional experiments and one real-world experiment involving three medical institutions were conducted to validate RevL, using three binary classification EHR data. In the real-world experiment with data from 106,508 patients, the mean global area under the receiver operating curve was 0.710 for RevL and 0.655 for TL. These results demonstrate RevL's ability to retain previously learned knowledge and its effectiveness in real-world PPDL scenarios. Our work establishes a realistic pipeline for PPDL research based on model transfers across institutions and highlights the practicality of continual learning in real-world medical settings using private EHR data.

Authors

  • Jaesung Yoo
    School of Electrical Engineering, Korea University, Seoul, Republic of Korea.
  • Sunghyuk Choi
    Department of Biomedical Engineering, Seoul National University College of Medicine, Republic of Korea.
  • Ye Seul Yang
    Department of Medicine, Seoul National University College of Medicine, Republic of Korea.
  • Suhyeon Kim
    Department of Applied Statistics, Chung-Ang University, Republic of Korea.
  • Jieun Choi
    Graduate School of Medical Science and Engineering, KAIST, Daejeon, Republic of Korea.
  • Dongkyeong Lim
    Department of Applied Statistics, Chung-Ang University, Republic of Korea.
  • Yaeji Lim
    Department of Applied Statistics, Chung-Ang University, Republic of Korea.
  • Hyung Joon Joo
    Department of Radiology (J.Y.L., Y.W.O., S.H.H.) and Division of Cardiology, Department of Internal Medicine (D.S.L., C.W.Y., J.H.P., H.J.J.), Korea University Anam Hospital, 73 Inchon-ro, Seongbuk-gu, Seoul 02841, Republic of Korea; Department of Radiology, Korea University Guro Hospital, Seoul, Republic of Korea (H.S.Y., E.Y.K.); and Department of Radiology, Korea University Ansan Hospital, Ansan, Republic of Korea (C.K., K.Y.L.).
  • Dae Jung Kim
    UMEDIX Co., Ltd., Seoul 06097, Korea. khpark@unist.ac.kr.
  • Rae Woong Park
    Department of Biomedical Informatics, Ajou University School of Medicine, Suwon, Gyeonggi-do, Republic of Korea.
  • Hyung-Jin Yoon
    Department of Human Systems Medicine, Seoul National University College of Medicine, Seoul, Korea.
  • Kwangsoo Kim
    Transdisciplinary Department of Medicine & Advanced Technology, Seoul National University Hospital, Seoul, Republic of Korea.