by Bill Snyder
Researchers at Case Western Reserve University and Vanderbilt University Medical Center are pioneering the use of computer-aided image analysis to improve the prognostication and treatment of head and neck cancer.
Each year in the United States, approximately 16,000 patients are diagnosed with a form of head-and-neck cancer called oropharyngeal squamous cell carcinoma (OPSCC) caused by the sexually transmitted human papilloma virus (HPV).
Not all cancers are alike, however, and some patients receive more of the standard treatment — aggressive chemotherapy and radiation — than is necessary to effectively treat their tumors.
“We have been overtreating many patients with chemotherapy and radiation that they do not need because we didn’t have a way to find out which patients would benefit from de-escalation,” said Anant Madabhushi, PhD, director of Case Western’s Center for Computational Imaging and Personal Diagnostics (CCIPD) in Cleveland.
In previous studies of this cancer, an increased density of tumor-infiltrating lymphocytes (TILs) in and around the tumor has been associated with lower risk of recurrence. In general, the more TILs, the less aggressive the cancer.
TILs are immune cells, some of which “are really good at catching early (tumor) cells that are growing out of control and neutralizing them,” explained James Lewis Jr., MD, professor of Pathology, Microbiology and Immunology and of Otolaryngology at VUMC.
It is difficult for pathologists looking through a microscope at slides of stained tissue samples to quantify the TILs or characterize their spatial patterns.
To meet this challenge, Madabhushi, Lewis and their colleagues digitized more than 1,000 slides collected from six hospital systems across the country, and they used machine learning techniques to “teach” the computer to recognize the patterns in a standardized way.
The result was an imaging biomarker they called OP-TIL, which successfully identified a subset of patients with less aggressive cancer who might have benefitted from a significantly reduced dose of radiation, and which was better at prognosis prediction than human visual review of the digital slides.
The findings were reported recently in the Journal of the National Cancer Institute. Lewis and Mitra Mehrad, MD, and Kim Ely, MD, both associate professors of Pathology, Microbiology and Immunology at VUMC, contributed to and are co-authors of the paper.
While combined chemotherapy and radiation is often curative in patients with HPV-associated OPSCC, side effects including dry mouth, fatigue, nausea, difficulty swallowing and hair loss can be significant. The prospect of safely reducing radiation therapy in a subset of patients, therefore, would mean reduced side effects as well.
“We have been able to visually assess patients’ tumors microscopically for a very long time, but now, with this technology, we can actually extract meaningful and reproducible information from the morphology for prognosis and prediction,” said Lewis, who has lent his expertise to Madabhushi’s computational imaging program for several years.
Lewis and Madabhushi are co-principal investigators of two major research project (R01) grants from the National Cancer Institute of the National Institutes of Health that are applying artificial intelligence (AI) and computer-aided image analysis techniques to improve the prognostication and treatment of head and neck cancers.
In the future, pathology slides will be digitized and analyzed by computer routinely, Lewis predicted. “There are so many applications for computer-aided image analysis,” he said. “You can do image analysis on pigmented skin lesions, heart transplant biopsies and gastric lymphoma. Those things are getting closer to clinical practice.”
The CCIPD, in the Case Western School of Engineering, is a leader in AI-driven precision medicine research.
In the current study, Lewis helped design the “classifiers,” where the computer was trained to recognize tumor cells and TILs. He also helped assemble the large, diverse cohort of patient slides for digitized images and corresponding clinical outcome data that were provided by contributors from across the country.
The research was supported by the following grants from the National Cancer Institute: CA202752, CA208236, CA216579, CA220581, CA249992, CA257612, CA239055, CA199374, CA248226, CA239055 and CA254566.
Other funding was provided by the National Heart, Lung and Blood Institute, the National Institute for Biomedical Imaging and Bioengineering, the National Center for Research Resources, and the U.S. Departments of Veterans Affairs and Defense.