Fred Hutchinson and three other cancer centers are creating a comprehensive AI model without compromising patient confidentiality.
1. In October, Seattle-based Fred Hutch Cancer Center spearheaded the "Cancer AI Alliance" with Boston-based Dana-Farber Cancer Institute, New York City-based Memorial Sloan Kettering Cancer Center, and the Sidney Kimmel Comprehensive Cancer Center and Whiting School of Engineering at Johns Hopkins in Baltimore. The alliance is funded by $40 million from AWS, Deloitte, Microsoft, NVIDIA and Slalom.
2. The alliance is focused on leveraging combined data from participating centers to identify opportunities for cancer care and research and collaborate on computing infrastructure. It hopes to produce its first insights by 2025.
3. The AI is trained using patient data, medical images, genome sequencing and EHR records from consenting patients to find more effective ways to treat and understand cancer. To create the most comprehensive AI, the four centers must share patient information to ensure the AI has the most diverse information it can, according to a Jan. 6 Fred Hutch news release.
This is where patient confidentiality issues arise.
4. Fred Hutch and its partners have found a way to adjust the AI model settings to offset privacy risks. Each center will receive a copy of the same overall model and train it on their patient data, adjusting the variable settings to make predictions more accurate. Once complete, the centers will send their model to a central location to update and improve it. In this process, the centers can share the models without sharing patient data itself.
5. All four models will be adjusted and combined before being sent back to the alliance partners. The centers will then run the model through their own patient data again to further train it. This process may repeat until it matches the cancers seen across the network.
"These models — trained on the comprehensive and diverse data of patient experiences at four cancer centers and potentially more institutions — will help researchers make better sense of the complex molecular interactions underlying tumor biology, disease progression, response to therapy and resistance to treatment," the release said.