A team of Stony Brook University-led researchers in New York created a method using deep learning digital pathology to map cancerous immune cell patters that may help guide new cancer therapies.
The study was published online April 3 in Cell Reports.
Pathologists conduct tissue diagnostic studies for nearly all cancer patients, a majority of which are stained with hematoxylin and eosin (H&E). However, digital pathology approaches can create a “computational stain,” allowing practitioners to visualize and quantify image features that go beyond traditional density-related tumor information, according to Joel Saltz, MD, PhD, chair of biomedical informatics at Stony Brook University, and colleagues.
“This paper demonstrates that we can now use deep learning methods such as artificial intelligence to extract and classify patterns of immune cells in ubiquitously obtained pathology studies, and to relate immune cell patterns to the many other types of cancer patient molecular and clinical data,” said Saltz, in a Stony Brook statement.
In this study, researchers used more than 5,000 pathology data images of 13 different cancer types on its two convolutional neural networks (CNNs). This data trained its computational staining method to identify tumor lymphocyte-infiltrated (TIL) regions in digitized H&E stained tissue specimens.
The method subsequently created TIL maps which allowed cancer specialists to create tumor-immune information from routinely gathered pathology slides.
Results showed the TIL maps were related to the molecular characterization of tumors and patient survival.
“In comparing TIL fraction identified via molecular methods to TIL maps derived from digital image analyses of H&E images, we found good but certainly not perfect agreement,” wrote Saltz et al.
The team went on to state perfect agreement is not expected, but the findings may have profound implications for cancer treatment.
“In a clinical setting, rapid and automated identification of the degree and nature of TIL infiltrate might be instrumental in determining whether options for immunotherapy should be explored or whether more detailed and costly immune diagnostics should be introduced,” wrote Saltz et al.
“This present study demonstrates value that can be added by careful examination of this rich resource, and it is our sincere hope that others will soon explore the many facets of these imaging data."