Open problems in medical federated learning

• DOI: https://doi.org/10.1108/IJWIS-04-2022-0080
• Published date: 20 September 2022
• Date: 20 September 2022
• Pages: 77-99
• Subject Matter: Information & knowledge management,Information & communications technology,Information systems,Library & information science,Information behaviour & retrieval,Metadata,Internet
• Author: Joo Hun Yoo,Hyejun Jeong,Jaehyeok Lee,Tai-Myoung Chung

Open problems in medical

federated learning

Joo Hun Yoo

Department of Artificial Intelligence, College of Computing and Informatics,

Sungkyunkwan University, Suwon, Republic of Korea, and

Hyejun Jeong,Jaehyeok Lee and Tai-Myoung Chung

Department of Computer Science and Engineering, College of Computing and

Informatics, Sungkyunkwan University, Suwon, Republic of Korea

Abstract

Purpose –This study aims to summarize the critical issues in medical federated learning and

applicable solutions. Also, detailed explanations of how federated learning techniques can be

applied to the medical field are presented. About 80 reference studies desc ribed in the field were

reviewed, and the federated learning frameworkcurrently being developed by the research team is

provided. This paper will help researchers to build an actual medical federated learning

environment.

Design/methodology/approach –Since machine learning techniquesemerged, more efficient analysis

was possible witha large amount of data. However, data regulationshave been tightened worldwide, and the

usage of centralized machinelearning methods has become almost infeasible. Federated learningtechniques

have been introduced as a solution. Even withits powerful structural advantages, there still exist unsolved

challenges in federatedlearning in a real medical data environment. This paper aims to summarizethose by

categoryand presents possible solutions.

Findings –This paper provides fourcritical categorized issues to be aware of when applying the federated

learning technique to the actual medical data environment, then provides general guidelines for building a

federatedlearning environment as a solution.

Originality/value –Existing studies have dealt with issues such as heterogeneity problems in the

federated learning environmentitself, but those were lacking on how these issues incur problems in actual

working tasks. Therefore, this paper helps researchers understand the federated learning issues through

examplesof actual medical machine learning environments.

Keywords Heterogeneity, Data security, Data privacy, Federated learning, Incentive mechanism,

Medical application

Paper type Research paper

© Joo Hun Yoo, Hyejun Jeong, Jaehyeok Lee and Tai-Myoung Chung. Published by Emerald

Publishing Limited. This article is published under the Creative Commons Attribution (CC BY 4.0)

licence. Anyone may reproduce, distribute, translate and create derivative works of this article (for

both commercial and non-commercial purposes), subject to full attribution to the original publication

and authors. The full terms of this licence may be seen at http://creativecommons.org/licences/by/4.0/

legalcode

This research is supported by Institute of Information & Communications Technology Planning &

Evaluation (IITP) grant funded by the Korea government(MSIT) (No.2019-0-00421, AI Graduate School

Support Program(Sungkyunkwan University)) and funded by Institute of Information & Communications

Technology Planning & Evaluation (IITP) grand funded by the Korea government(MSIT)(No.2020-0-00990,

Platform Development and Proof High Trust & Low Latency Processing for Heterogeneous·Atypical·Large

Scaled Data in 5G-IoT Environment).

Medical

federated

learning

77

Received15 April 2022

Revised14 June 2022

Accepted14 June 2022

InternationalJournal of Web

InformationSystems

Vol.18 No. 2/3, 2022

pp. 77-99

EmeraldPublishing Limited

1744-0084

DOI 10.1108/IJWIS-04-2022-0080

The current issue and full text archive of this journal is available on Emerald Insight at:

https://www.emerald.com/insight/1744-0084.htm

1. Introduction

Machine learning has been widely studied in various research fields for its powerful

performance in data analysis. It was possible to derivebetter results through machine learning

methods by learning the hidden multi-dimensional characteristics of given data that were

difficult for humans to distinguish. This structure of machine learning in the medical imaging

field, where it is crucial to capture fine features in images, has been very helpful in

strengthening the existing diagnostic approaches. For example, support vector machines, deep

neural networks, convolutions and clustering techniques have been applied in the medical field

to effectively search those human-unidentifiable correlations from medical data.

Through the active use of machine learning approaches, the medical field was able to expand

its scope to specific medical fields such as radiology, pathology, neuroscience, genetics and even

mental disorders. However, the biggest issue in the field of medical artificial intelligence (AI) is

not the accuracy of diagnosis, but the protection of patients’personal information.

Federated learning, a machine learning algorithm based on the distributed data

environment, has emerged under stricter data regulations laws around the world. When the

concept of federated learning was first introduced, data privacy regulations such as the EU’s

General Data Protection Regulation, California’s(CA’s) Privacy Rights Act and China’s

Personal Information Protection were representative rules, but now more countries around the

world are implementing efficient regulations, such as Brazil’s Lei Geral de Prote,c~ao de Dados,

Canada’s Digital Charter Implementation and Singapore’s Personal Data Protection Act, to

protect their citizens’personal information. Thus, centralized machine learning methods, that

collect and learn based on the proper amount of data, are no longer applicable under the

personal data protection regulations. In particular, for medical data, researchers and business

providers should follow the Health Insurance Portability and Accountability Act (HIPAA),

which comprehensively protects the medical records and independently identifiable health

information of patients and medical information providers. With numerous increasing data

regulations, researchers have applied various solutions to prevent invasion of privacy.

First, the most common solutions to adopt for data privacy issues are to process and

import variables that can identify individual users when collecting their data. Primary

information leakage can be prevented through the measures such as secure aggregation,

pseudonymization, data reduction, data suppression and data masking in normal data

environments. However,personal health information (PHI) is difficult to apply these security

methods, as it contains any format of information thatcan identify the data owner. PHI is a

wider concept of personal identifiable information (PII), which means sensitive information

such as health insurance records, medical numbers, health status, medical images and

mental health records are included on top of basic user variables. Therefore, data security

methods for PHI are difficult to completely protect the information and to use medical data

efficiently and safely,structural solutions are required to learn without violations.

Federated learning is a structural solution for the existing data privacy violation

problems of machine learning methods. It has a unique structure and characteristics

compared to centralized machinelearning. Traditional machine learning approaches require

a large volume of training data collected from local data owners to the server for model

generation. Federated learning, a decentralized learning structure, generates and develops

deep neural network models without local data collection to the server. The core concepts

used in the neural networkmodel learning process are as below.

Individual clients’data stored in the local environment does not move, and the server

generates the initial training model and delivers it to each participating client. The transferred

initial training model goes through a model update process through learning within each

client’s data environment, and the server collects all of the corresponding results to create a

IJWIS

18,2/3

78