5 Insights on Leveraging AI and Imaging Data to Drive Oncology Research
Oncology research often relies on medical imaging data to develop new therapies. Medical imaging has the power to reveal patterns in diagnosis and the effectiveness of novel therapies, and artificial intelligence (AI) can boost human capabilities greatly, in this regard. But it can be challenging to corral enough imaging data, in the same format, with adequate metadata attached to develop useful AI algorithms.
In a recent discussion with Becker’s Hospital Review, “Leveraging imaging data to drive advancements in oncology: An expert discussion,” we spoke with Despina Kontos, Ph.D., Professor of Radiology, Vice Chair of AI and Data Science Research, and Director of the Center for Innovation in Imaging Biomarkers and Integrated Diagnostics (CIMBID) at Columbia University Irving Medical Center. Read on to learn real-world insights on challenges researchers face in leveraging AI and imaging data — and how to move past them.
AI can vastly multiply what can be identified by the human eye
Researchers like Dr. Kontos are seeking to use advanced computing, AI and machine learning to extract thousands of measurements from imaging data.
“[These measurements] can give us additional information than what we can just capture with our bare eye or with more conventional measurement,” Dr. Kontos explains.
For example, when looking at mammograms, a machine learning approach can help identify an area within similar images from which you can extract imaging data and get information about the entire breast tissue, rather than only breast density. This data may show information about heterogeneity, the contrast in the images, different patterns that may exist in the tissues, and so forth, far beyond what can be achieved by human review alone.
Valuable information may be sitting in your repository
Imaging files can hold a wealth of valuable data — but it’s difficult to access all of that information without the help that AI can provide.
“From routinely collected imaging data, data that is sort of sitting in your PACS system and nobody is ever accessing it after the original interpretation, we can extract additional information relatively easier with computer science and AI algorithms that have additional prognostic value,” Dr. Kontos explains.
Data privacy is a surmountable concern
One of the most pressing concerns in the field is data privacy, especially when dealing with sensitive medical imaging data. Dr. Kontos acknowledged these concerns but pointed out that technological advancements exist to help researchers maintain privacy.
“There are robust technologies in place to de-identify the data to remove HIPAA concerns,” Dr. Kontos says. “There are ways to do compliance agreements between institutions, and there are also HIPAA-compliant computing and storage platforms.”
With the right frameworks in place, privacy challenges can be effectively managed, allowing researchers to focus on the scientific and technological aspects of their work.
Technology can help connect research to clinical practice
Advancements made in oncology research should translate into tangible benefits for patients. But the journey from research to clinical practice is fraught with bottlenecks, such as a lack of available data and an inability to organize that data.
“How do we store, how do we organize, how do we index this data, how do we link this data with metadata ... and how do we bring all this information together?” Dr. Kontos says.
Without a high enough volume of diverse patient data, findings from clinical trials may not be useful — the same holds true for AI models trained on insufficient data.
“You can’t and shouldn’t give a tool that was trained on 300 patients to a neurosurgeon,” Dr. Kontos says. “That’s not okay.”
A solution like Flywheel enables researchers to securely manage and share data from various locations and institutions, as well as access additional publicly available datasets — so that AI models developed from that data will be usable on more patient populations.
Variability in medical imaging data can be overcome
The biggest challenge in using medical imaging data for AI development is the variability in reporting between radiologists and across institutions. This variability can hinder the development of AI models that generalize well across different settings. Additionally, the diversity of data sets poses another challenge.
“How do you know if an AI model is going to work at your institution if it’s been developed at another?” Dr. Kontos notes.
However, these challenges also present opportunities. Platforms like Flywheel offer solutions by allowing researchers to build models using their own data or test and validate externally developed algorithms on their internal data. This capability is crucial for developing robust AI models that can be applied broadly, ensuring that advancements in AI benefit a wide range of patients.
Moving forward with support from Flywheel
Having the right tools and infrastructure to overcome the challenges associated with using imaging data in oncology research is crucial. Platforms like Flywheel can play a pivotal role in this by providing the necessary infrastructure to manage, analyze and share imaging data efficiently and securely.
To learn more about how to leverage imaging data to drive advancements in oncology, we encourage you to watch our full conversation with Dr. Kontos. If you’re ready to take the next step in your research, get in touch with Flywheel to learn how our medical imaging platform can help you make new discoveries in the battle against cancer and improve patient outcomes.
