NCI-Funded Tool Combines Imaging Modalities to Aid Pathologists and Machine Learning (ML)
Are you developing technology for cancer research and wondering how NCI’s Small Business Innovation Research (SBIR) grant might help? In this study, NCI-funded researchers are using an SBIR grant to develop and test a tissue imaging technology, called “Orion.”
Read the full report, “High-Plex Immunofluorescence Imaging and Traditional Histology of the Same Tissue Section for Discovering Image-Based Biomarkers,” in Nature Cancer. You can access the code on Zenodo.
Orion lets you interpret whole-slide images of tissue sections (mainly mouse or human), blending hematoxylin and eosin (H&E) with an imaging technique called immunofluorescence (IF), a method that tags key molecular features so they’re easy to distinguish. With a single digital slide, you can see cellular characteristics and tissue architecture, as well as distinguish cell activity.
In short, the platform lets you measure specific molecular markers (using IF) along with the morphological features that pathologists have used for centuries to diagnose and stage cancer (using H&E)—in the same cells.
By combining these two imaging modalities, Orion offers insight into immune activity. According to corresponding author, Dr. Peter Sorger of Harvard Medical School, “Measuring the extent and type of immune response in cancer is vital for predicting therapeutic response and outcomes, which are critical components of precision medicine.”
Thanks to SBIR support, Orion is commercially available and the analysis code is open source, making the technology easy to access.
As noted by Dr. Sorger, “Orion data will be useful for training the machine learning (ML) algorithms used in digital pathology so they can better classify cell types and states that aren’t easily identified using H&E images alone.”
He added, “Multi-modal data and ML models combining H&E and IF are certain to be more beneficial than either modality used alone. In a research setting, multi-modal analysis gives you tissue-level context for single-cell spatial profiling. In a diagnostic setting, multimodal ML models can help identify and explain the molecular features underlying cancer and its progression.”