Federated Machine Learning Gives Healthcare Organizations a Competitive AI Advantage

Real-World Use Cases in Healthcare

There is a bevy of use cases for which organizations in a range of industries can use federated ML, Lange notes. In healthcare, hospitals collaboratively train models for cancer diagnosis, brain tumor segmentation and COVID-19 detection without sharing patient records, she says. For example, U.S. medical centers — including collaborators from Case Western Reserve University; Georgetown University; Mayo Clinic; the University of California, San Diego; the University of Florida; and Vanderbilt University — are using NVIDIA-powered federated learning for tumor segmentation, according to an NVIDIA blog post.

Key Infrastructure Requirements for Federated Learning Deployments

To run federated learning, healthcare organizations need a central system to coordinate the process (including model distribution, scheduling and update aggregation), local infrastructure with sufficient computing power to train models and secure ways to send updates between participants, according to Lange.

Also Read

Study describes three distinct trajectories of cognitive decline in Alzheimer’s disease

Centre asks states to step up fire safety in healthcare facilities

MedTech sees trade upside in India–NZ FTA, but missing regulatory recognition a concern

Plugin developed by ProSEOBlogger. Get free gpl themes.

Google notes in its blog that organizations must bring the model to client devices to perform the local model training, and these can range from mobile phones to IoT devices and entire institutions, such as hospitals. 

The central server or aggregator “acts as the orchestrator of the federated learning process,” Google notes. “It initializes and distributes the global model, collects model updates from clients, aggregates these updates to refine the global model, and then redistributes the updated model. It doesn’t directly access the clients’ raw data.”

READ MORE: Build a resilient AI ecosystem in healthcare beyond compliance.

IT leaders also need to use a defined communication protocol to determine how “clients and the server exchange information, primarily the model parameters and updates. Efficient and secure communication protocols are crucial, especially given the potential for a massive number of clients and varying network conditions.” 

Finally, a model aggregation algorithm is how the central server combines the model updates received from the clients. “Algorithms like federated averaging are commonly used to average the weights or gradients, creating a single, improved global model,” Google notes.

Organizations also need data and model governance, says Lange.

Building a Cross-Enterprise AI Ecosystem: The Key Takeaway

Ultimately, federated learning is about cross-enterprise collaboration for building AI models without sharing raw data, Lange says. 

“In regulated, multienterprise environments, it can unlock better models without forcing organizations to give up control over their most sensitive data,” she adds. “For success, it’s critical that the participants establish clear agreements on data ownership, contributions and responsibilities.”