Infrastructure platform for privacy-preserving distributed machine learning development of computer-assisted...

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Data-driven clinical decision support tools are increasingly used in cancer care.
• Challenges in data sharing (privacy, administrative, political barriers) hinder the development and scope of these tools.
• Federated learning offers a privacy-preserving approach for utilizing distributed datasets.

Purpose of the Study:

• To develop and implement an IT infrastructure for privacy-preserving clinical decision support systems using federated learning.
• To harmonize routinely collected clinical and imaging data across multiple institutions.
• To demonstrate the infrastructure's utility in developing prognostic models for radiation oncology.

Main Methods:

• Developed and deployed a web service software infrastructure enabling machine learning on geographically distributed datasets.
• Established harmonized, ontology-linked lung cancer databases across Australian hospitals.
• Utilized federated learning to build logistic regression models predicting cardiovascular events post-radiation therapy.

Main Results:

• The Australian computer-assisted theragnostics (AusCAT) network was established, involving four radiation oncology departments.
• A model predicting cardiovascular admission within a year of curative radiotherapy for non-small cell lung cancer was developed.
• The logistic regression model achieved an AUROC of 0.70 and C-index of 0.65 on out-of-sample data.

Conclusions:

• The developed infrastructure is practical for inter-clinical data harmonization and model development using federated learning.
• This approach shows promise for advancing radiation oncology research using real-world clinical data.
• Federated learning facilitates collaborative research while preserving patient privacy across distributed healthcare networks.