Predictive models are gaining ground across healthcare as more and more hospitals and specialists use them for diagnosis and treatment of cancer and other diseases. But these machine learning tools are still not as accurate or powerful as they could be – and that often boils down to not having enough quality clinical data on which to be trained.
One way to help address the fact that many sample sizes are just too small is to aggregate data from other sources, says Steve Irvine, founder and CEO of integrate.ai. That can be done, while protecting patient privacy, with federated learning techniques, which can open up vast new troves of data for researchers. He explains more in this episode of HIMSSCast.
Like what you hear? Subscribe to the podcast onApple Podcasts,Spotify orGoogle Play!
Talking points:
How machine learning models for oncology have evolved in recent years
The keys to building a good predictive algorithm
Why finding enough quality data to train models is so challenging
What federated learning is, and how it can help
Opportunities and challenges of embracing a federated learning approach
How Irvine sees predictive oncology models evolving in the years ahead
More about this episode:
Top 10 AI and machine learning stories of 2022
HIMSSCast: 2023 forecast - 5G, AI command centers, hybrid work models and more
Demystifying AI's role in healthcare to reassure new providers – and old pros
AI is fast addressing data requirements and advancing interoperability, says one expert
Oncology practice uses AI to significantly improve end-of-life care
How one health system developed AI to tackle patient safety
Orlando Health to launch AI-driven hospital-at-home services