Leveraging Digital Pathology and Deep Learning to Predict Immunotherapy Response
Abstract
Traditionally, patients with metastatic clear cell renal cell carcinoma have been treated with therapies that targeted the tumor blood vessels. More recently, immune-based therapies have emerged that have significantly changed the treatment and outcomes for our patients. Immune checkpoint inhibitors (ICIs) are drugs that use the body s immune system to fight diseases such as cancer. Today, patients with metastatic renal cell carcinoma are treated with combinations of drugs that involve ICI and/or drugs that target the tumor blood vessels. However, none of these treatments uniformly benefit all patients and many suffer from a lot of side-effects from these drugs. Different drugs target different molecular pathways and possibly benefit different groups of patients and, therefore, there is a need for markers that can tell us if the patient will respond to one group of drugs or not (ICI or those that target blood vessels). These markers need to be easy to incorporate into everyday practice. Past studies have tried to identify patients that may respond to ICIs by correlating levels of a protein called PD-L1 on tumors to how patients respond to treatment. These studies have found only a minimal correlation between clinical outcomes and PD-L1 levels. Patients with no PD-L1 expression may also respond. Recently, data from a clinical trial showed that a set of gene expression signatures could predict if patients would be susceptible to drugs that target blood vessels or respond better to a combination drug regimen that included ICI. These results have now been validated in an independent large clinical trial. In this proposal, we will use artificial intelligence and apply it to the digital images from routinely used tumor sections to predict the gene expression signature. This will allow us to evaluate the gene expression without the need for expensive and time-consuming studies like sequencing. Our trained computers can assess the tumor blood vessels and inflammation in the tumor and thereby predict responses to these therapies that target the blood vessels or the inflammatory cells. If successful, our efforts will lead to the first predictive biomarker in renal cell carcinoma that can be easily incorporated into everyday clinic. This will enable appropriate allocation of drugs to the patients who will show treatment benefit and not be given to those patients who will not benefit and thus minimize toxicity.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 05, 2021
- Source ID
- W81XWH2110630
Entities
People
- Payal Kapur
Organizations
- United States Army
- University of Texas Southwestern Medical Center