Acquiring clinical data to develop effective AI tools for medical imaging comes with a number of challenges, including patient privacy, partnerships and legal and ethical requirements
Radiologists discuss data sharing practices for AI development at RSNA
December 09, 2020
by John R. Fischer
, Senior Reporter
Acquiring and sharing clinical data to develop AI tools comes with a number of challenges, from addressing concerns over patient privacy to the way in which data is shared.
"We need to optimize protocols to minimize the risks of data. We need validated and frankly better methods of deidentification, and we need to investigate safer means of data sharing,” said Dr. Yvonne Lui, associate chair of artificial intelligence in the department of radiology at NYULH and a radiology professor at NYU School of Medicine, during a virtual session at the 2020 RSNA Annual Meeting.
She warns that certain practices, such as skull stripping, can alter the quality and accuracy of data used to develop AI. She also encouraged radiologists to work with manufacturers to avoid exposing identifiable information such as when it is placed in proprietary DICOM fields, and discussed the issues of HIPAA, which does not cover and protect certain forms of data, including research records, aggregated information from startup and technology companies, and medical devices.
“There is talk about whether we need to revise HIPAA. Do its 18 identifiers need to be expanded? Do we need to entirely re-envision HIPAA and make it less constrictive,” said Lui.
Another challenge can be academic industrial partnerships between institutions and companies, according to Dr. Julius Chapiro, assistant professor of radiology and biomedical imaging at the Yale University School of Medicine. These consist of M.D.s, Ph.D.s, engineers and computer scientists who work together to develop products such as AI solutions for medical imaging.
Chapiro says such a collaboration requires each side to understand what the other brings to the table and legalities around their work, such as protocols on profit and ownership rights. “Choose the right idea and identify the right data. Be aware of local and legal frameworks and stay on the safe side both from a legal and ethical perspective. Be aware of AI hypes. Don’t overpromise and underdeliver.”
Sharing information of any kind in healthcare also requires all healthcare players to understand their obligations in disseminating data and work together. “Everyone who participates in the healthcare system has a number of obligations to make the system better, to respect the people who are in it, respect judgements and address health inequities,” said Dr. David Lawson of Stanford University School of Medicine in reference to the 2013 Hastings Center Report: An Ethics Framework for a Learning Healthcare System.
This includes patients, he adds, who have an obligation to help improve the quality and value of clinical care and the healthcare system. “Often that contribution is nothing more than being part of the healthcare system and allowing individuals and organizations to learn from them if they do so in an ethical manner.”