Collaborative Privacy-preserving Approaches for Distributed Deep Learning Using Multi-Institutional Data.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Radiology

Background:

• Deep learning (DL) shows promise in medical imaging but struggles with generalizability due to limited, single-institution data.
• Centralized data pooling for DL training raises privacy, cost, and regulatory concerns.

Purpose of the Study:

• To explore distributed machine learning techniques for training DL models on multi-institutional medical data.
• To address the challenges of data privacy and logistical hurdles in collaborative medical AI development.

Main Methods:

• Review of popular collaborative training methods in distributed deep learning.
• Discussion of deployment considerations and available software frameworks for federated learning.
• Showcasing real-world examples of collaborative learning in medical AI.

Main Results:

• Distributed learning facilitates robust DL model development without centralizing sensitive patient data.
• Federated learning and other distributed approaches offer viable solutions for multi-institutional medical AI training.
• Identified key challenges and future research directions for distributed DL in healthcare.

Conclusions:

• Distributed DL is crucial for enhancing the generalizability and robustness of medical AI algorithms.
• Clinicians should understand the benefits, limitations, and risks associated with distributed DL.
• Further research is needed to overcome existing challenges and advance distributed DL in clinical practice.