Abstract:In this paper, we compare three privacy-preserving distributed learning techniques: federated learning, split learning, and SplitFed. We use these techniques to develop binary classification models for detecting tuberculosis from chest X-rays and compare them in terms of classification performance, communication and computational costs, and training time. We propose a novel distributed learning architecture called SplitFedv3, which performs better than split learning and SplitFedv2 in our experiments. We also propose alternate mini-batch training, a new training technique for split learning, that performs better than alternate client training, where clients take turns to train a model.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the challenges encountered when applying privacy - protected distributed deep - learning methods in the healthcare field. Specifically, the paper aims to compare three privacy - protected distributed learning techniques: Federated Learning (FL), Split Learning (SL) and SplitFed Learning, and evaluate their performance, communication and computational costs, and training time in detecting tuberculosis (TB) in chest X - rays. ### Summary of Main Problems: 1. **Data Privacy and Compliance**: - Medical data are usually distributed among different medical institutions and are strictly protected by laws and regulations (such as GDPR and HIPAA) and cannot be freely shared. Therefore, how to use these data for model training without violating privacy regulations is a key issue. 2. **Effectiveness of Distributed Learning**: - By comparing the three distributed learning methods of FL, SL and SplitFed, the paper explores their effectiveness and feasibility in practical applications. In particular, the paper proposes a new distributed learning architecture, SplitFedv3, and introduces a new training technique - alternate mini - batch training - to improve model performance. 3. **Performance Evaluation**: - The paper not only focuses on classification performance (such as AUROC, AUPRC, F1 - score and Kappa coefficient), but also considers important factors in practical applications such as training time, data communication volume and computational cost to comprehensively evaluate the advantages and disadvantages of different methods. ### Specific Research Contents: - **Dataset**: Use chest X - ray image datasets from five different sources, including three private datasets and two public datasets (MIMIC and Padchest). - **Model Architecture**: The experiment uses two neural network architectures, DenseNet - 121 and U - Net, and tests them at two different resolutions respectively. - **Experimental Setup**: Describe in detail the specific configurations and training processes of each distributed learning method, including the federated averaging algorithm in federated learning, different configurations in split learning (label - sharing and non - label - sharing), and different versions of SplitFed (SFLv2 and SFLv3). Through these studies, the paper provides a valuable reference for distributed deep learning in the medical field and points out the key factors that need to be considered in practical applications.

Comparison of Privacy-Preserving Distributed Deep Learning Methods in Healthcare

Split Learning for Distributed Collaborative Training of Deep Learning Models in Health Informatics

Distributed Learning Approaches for Automated Chest X-Ray Diagnosis

Split Learning for collaborative deep learning in healthcare

Systematic Review of Privacy-Preserving Distributed Machine Learning From Federated Databases in Health Care

Differential Privacy-enabled Federated Learning for Sensitive Health Data

Introduction and Comparison of Novel Decentral Learning Schemes with Multiple Data Pools for Privacy-Preserving ECG Classification

Advancements of federated learning towards privacy preservation: from federated learning to split learning

A Study of Federated Learning with Internet of Things for Data Privacy and Security using Privacy Preserving Techniques

DPCOVID: Privacy-Preserving Federated Covid-19 Detection

A Review of Privacy Enhancement Methods for Federated Learning in Healthcare Systems

Privacy-Preserving Breast Cancer Classification: A Federated Transfer Learning Approach

SplitAVG: A Heterogeneity-Aware Federated Deep Learning Method for Medical Imaging

Privacy preserving distributed machine learning with federated learning

Secure Federated Learning Approaches to Diagnosing COVID-19

Vision Through the Veil: Differential Privacy in Federated Learning for Medical Image Classification

Institutionally Distributed Deep Learning Networks

Split learning for health: Distributed deep learning without sharing raw patient data

Privacy-Preserving Decentralized Learning Methods for Biomedical Applications

[Plasma secretin concentration and pancreatic secretion].

Federated Learning with Deep Neural Networks: A Privacy-Preserving Approach to Enhanced ECG Classification.