Penn Medicine develops AI tool for precision oncology
The application, known as iStar, was created by researchers at the Perelman School of Medicine to give doctors greater insight into gene activity in medical images and potentially help them diagnose cancers that might otherwise go unnoticed.
Researchers at Penn Medicine have developed a new artificial intelligence application that provides a new way to examine and interpret medical images and could help doctors diagnose cancers that may not have been previously discovered.
WHY IT MATTERS
The new tool, called iStar – it stands for Inferring Super-Resolution Tissue Architecture – was created at U Penn's Perelman School of Medicine. The computing power enables detailed views of individual cells in images, and could help oncologists and researchers see cancer cells that might otherwise have gone unnoticed.
As explained in a recent Nature paperAccording to Penn Medicine, the AI tool can help determine whether safe margins have been achieved after cancer surgery, and can also provide automatic annotation for microscopic images, enabling new developments in the diagnosis of molecular diseases.
The iStar technology was developed from National Institutes of Health-funded research led by Mingyao Li, professor of biostatistics and digital pathology at the Perelman School, and Penn Medicine research associate David Zhang.
The application can automatically detect critical anti-tumor immune formations known as tertiary lymphoid structures. The presence of these formations correlates with a patient's likely survival and favorable response to immunotherapy, Li said, indicating that iStar could be immensely helpful in determining whether a particular patient would benefit from specific immunotherapy interventions.
Penn Medicine notes that iStar's research and development stems from the emerging field spatial transcriptomics, which maps gene activities in tissue space. By adapting a machine learning tool called the Hierarchical Vision Transformer, Li and her colleagues trained it on standard tissue images.
Starting by segmenting images into different stages — starting with looking for fine details, then moving up and “understanding broader tissue patterns,” as Li explained — the iStar AI uses that data in context with other clinical information, and applies it to predict gene activities, often with near-single-cell resolution.
Li and her colleagues tested the tool by evaluating iStar on different types of cancerous tissue, in addition to healthy tissues. In those tests, the technology was able to “automatically detect tumor and cancer cells that were difficult to identify by eye,” according to Penn Medicine, which noted that “physicians may be able to diagnose and diagnose more difficult conditions in the future.” visible or difficult to identify cancers, where iStar acts as a support layer.”
THE BIG TREND
Artificial intelligence is enabling major advances in more personalized and patient-centered care – just as innovative policies and more powerful computers are paving the way for further innovation in precision medicine and genomic programs and other AI-based oncology treatments.
ON THE RECORD
“iStar's power comes from its advanced techniques, which inversely mirror how a pathologist would study a tissue sample,” Li said in a statement. “Just as a pathologist identifies broader areas and then zooms in on detailed cellular structures, iStar can capture the overarching tissue structures and also focus on the details in a tissue image.
Furthermore, she noted, iStar can be applied to a significant number of samples – a key necessity for large-scale biomedical studies.
“The speed is also important for current expansions in 3D and biobank sample prediction,” says Li. “In the 3D context, a tissue block can comprise hundreds to thousands of serially cut tissue sections. The speed of iStar makes it possible to reconstruct this enormous amount of spatial data in a short time.”
Mike Miliard is editor-in-chief of Healthcare IT News
Email the writer: mike.miliard@himssmedia.com
Healthcare IT News is a HIMSS publication.