Scientists at UW-Madison have detailed a way to incorporate numerous forms of medical data into a set of machine learning models, in order to better predict the course of diseases such as cancer.
WARF included the method in its latest round-up of top licensing prospects within its healthcare portfolio, noting it offers a “multi-scale approach to information analysis” across distinct datasets, ranging from medical scans to complex genetic information.
In their patent application, inventors Prof. Pallavi Tiwari and student Olivia Krebs noted routine cancer treatments involve bringing together complementary data sources such as CT and MRI images, results from genome sequencing, cellular-scale findings and clinical data such as age and cognitive scores.
“Comprehensive assessment of this deluge of data by clinicians alone is infeasible,” they wrote, adding “it would be valuable to develop computational frameworks that integrate complementary cross-scale information … to enable discovery of new markers that can pave the way for precision medicine.”
They describe the potential for “integrated prognostic markers” to reliably predict risks associated with survival, as well as disease recurrence or progression.
In its overview on the technology, WARF authors note both diagnosis and prognosis in healthcare stands to benefit from the “recent increase in the performance, costs, and access to sophisticated computational resources” for complex analysis.
Though earlier computational models drawing together various forms of health data have shown improvement over older approaches, the inventors say issues with these efforts are still largely unaddressed. Those include retrospective datasets being unavailable and the challenge of carefully optimizing strategies for combining information, they wrote.
Researchers argue earlier approaches to this challenge “do not take full advantage of complementary information and complex multi-scale, cross-correlations across and between” the diverse datasets that are now available.
Their systems and methods involve establishing a hierarchical framework for assessing key factors for a given condition, wrapping in measurements at various scales of size and frames of reference. The approach can then ultimately “capture long-range and complex associations,” offering new insights about how they may influence one another.
See the overview from the Wisconsin Alumni Research Foundation.
