Implementation of a scalable platform to enable regulated digital health product development and commercialisation

Start Date

24 August 2022

End Date

31 October 2025

Project Participants

Peter MacCallum Cancer Centre, Propel Health AI, Swinburne University of Technology

Tags: AI governance, artificial intelligence (AI), data linkage, health data infrastructure, multimodal data, radiology, research infrastructure, responsible AI

Opportunity

Healthcare organisations generate and possess large amounts of clinical, imaging, pathology and genomic data, but these datasets are often stored across disconnected systems. This limits their use in research and clinical innovation such as developing new digital health technologies.

The project sought to establish an integrated, interoperable and secure scalable platform which could harmonise multimodal healthcare data for data-driven research and development. The project also looked to create a use case to validate platform efficacy.

Project Objectives

This project aimed to establish the Propel Health AI (Propel) platform, utilising Peter MacCallum Cancer Centre (Peter Mac)’s data to demonstrate how secure, governed access to multimodal clinical data can unlock the value of clinical datasets, support collaborative research in digital health, and establish a pipeline for the development of prototype decision support solutions.

The project also looked to demonstrate the platform’s AI and collaboration capabilities through the development of a multimodal vision-language foundation model as the basis for a prototype radiological reporting tool.

Publications

Written Conference PaperPublished 1st May 2026A Lightweight Large Vision-Language Model for Multimodal Medical Images
Full Publication
End-user PublicationPublished 1st June 2025Explainer: Artificial Intelligence for Digital Health
Full Publication

Final Outcomes

The project successfully established Propel’s health data platform, bringing together select datasets from different clinical data and imaging systems with appropriate governance and regulatory oversight, linking and harmonising the de-identified data for interoperability.

The platform was then used to develop and test RadAssist, a proof-of-concept AI co-pilot for improving existing practice and care. The lightweight large vision and language model (LVLM) prototype, which was trained and tested on Swinburne University of Technology’s Supercomputer, aims to help radiologists in report writing and clinical communications.

Impact and Benefits

The project provides a foundation for future digital health product development by showcasing the capabilities of a secure scalable platform for unlocking health data to advance AI innovation while maintaining strong privacy and governance controls. The proof-of-concept tool demonstrates the platform’s ability to enable collaborative development of efficient, lightweight AI models that can support real-world clinical workflows. The project also created valuable knowledge relating to AI governance, trust, ethics and implementation in healthcare settings, helping accelerate the adoption of responsible AI into clinical practice.