What if some of the leading organizations in technology, research and medicine worked together to share data and advance lifesaving science?
That vision is starting to take shape, with three pilot awards announced this fall by the Cascadia Data Alliance, launched in 2019 to create a Pacific Northwest data-sharing ecosystem comprising some of the region’s powerhouse institutions, working together across disciplines to accelerate innovation and save lives from cancer and infectious diseases.
The three Cascadia Collaboration Awards to cross-institutional teams at the Alliance’s member organizations — Fred Hutchinson Cancer Research Center, the University of Washington eScience Institute, BC Cancer, the University of British Columbia Data Science Institute and the Knight Cancer Institute at Oregon Health & Science University — represent more than $1.2 million in funding and credits for Microsoft’s Azure cloud computing service.
These projects are made possible thanks to support from Microsoft and institutional support from each participating organization.
The early stage funding aims to promote collaborations that may answer important scientific questions and to develop new ways for using technical solutions and best practices, data and methods standardization, and Azure cloud services that could be broadly applied in future research. The three projects tackle a range of questions in the cancer field.
Researcher from the Gynecologic Cancer Initiative are working on a project using machine learning to help diagnose specific cancer types.
As targeted treatments have become available for specific types of ovarian cancer, it’s become more important than ever that each patient receives an accurate diagnosis of her tumor type. To help make this possible, a research team wants to establish an international network for AI-based, privacy-protected pathology quality assurance. Ovarian cancer will be the team’s proof of concept for a system that could eventually be used for a variety of cancers.
The team is led by Fred Hutch’s Dr. Holly Harris, BC Cancer’s Dr. David Huntsman, OHSU’s Dr. Terry Morgan and Dr. Ivan Beschastnikh, a computer scientist at UBC. It also includes Simon Fraser University’s Dr. Tania Bubela and UBC’s Dr. Ali Bashashati, who developed the original algorithm the group hopes to generalize.
“A group of us at UBC has been working towards establishing a usable, end-to-end, privacy-enhancing system called LEAP, co-led by [UBC Professor of Medicine] Aline Talhouk and me” — said Beschastnikh — “that can be readily deployed in a health care setting. This project is a proof of concept for our ideas and, if successful, has the potential to impact how data is shared and analyzed in medicine.”
When someone is diagnosed with a cancer, a specialist called a pathologist studies a sample of their tissue, typically looking at the tissues under a microscope, to classify its type based on its biological features. These findings help to guide the treatments the patient receives.
The team plans to use real cancer pathology images from collaborating cancer centers to train an AI with a type of machine learning, or ML, to improve the pathological classification of ovarian cancer while ensuring patient privacy and data security.
“We will also collect privacy and security requirements informed by actual threats to health data,” Beschastnikh said. “This will help us to prepare our platform for implementation by striking the right balance between accuracy of results, privacy and security.”
He explained that their system will use a methodology called federated machine learning, which can train models without moving any real data to ensure it remains private.
“A key concern for modern ML/AI systems is privacy. This is because training high-quality models requires access to a lot of information, and in the medical domain this data is not only regarded as very private, it is also strictly regulated. Another concern is the fact that models trained on data remember aspects of the data and can even be used to reconstruct some of the original input data,” he said. The team’s federated ML methods are designed to avoid those problems, thus preserving patient privacy.
Besides identifying and counteracting privacy threats, the team will develop technical and socioethical guidelines for using the classifier worldwide. The team’s long-term vision is to establish a network of such AI-based systems that could be used by doctors everywhere, thereby ensuring more patients — even those far away from specialized cancer centers — receive accurate diagnoses and thus the most appropriate treatments for their specific cancer types.
“Ethical data sharing is the missing link to enable ML/AI applications in the health care domain. This project is at the intersection of health, machine learning, privacy, security and ethics,” said Beschastinikh.
Click here to read the full article on this research project.