Machine Learning-Based Integrated Morphological and Molecular Characterization of Advanced Prostate Cancer
Abstract
All prostate cancers are diagnosed by a pathologist, who evaluates and identifies the appearance of cancer cells under a microscope. The Gleason Grading System is widely used to determine the risk of a patient to develop disease progression and is the single most important information that is used to guide therapy after initial diagnosis. Although the Gleason Grading System is extremely powerful for assessing primary prostate cancer, it cannot be applied to metastatic prostate cancer. Thousands of men undergo biopsies of cancer metastases each year and, although the information content of these tumor tissues is very high, very few details derived from the analysis of these biopsies are currently used to inform clinical decisions. Following subjective review by pathologists, tissue slides are sent to physical storage where they are generally not re-accessed. An exceptional amount of information is lost in this process. As more and more patients undergo biopsies of metastatic prostate cancer, there is an urgent need to develop new approaches that allow us to extract the most import information from these biopsy tissues to best inform clinical management. Over the past years, advanced computer-based image and pattern analyses have been introduced into medicine. This technology is based on computational machine learning models that apply similar interconnected networks as used in the human brain, referred to as neural networks. These approaches are extremely efficient in detection and classification of objects (i.e., cancer) within very large complex images, tasks that can be very time-consuming for clinicians. Furthermore, they are powerful in detecting patterns that were previously not apparent to doctors and researchers due to their ability to combine data from multiple sources, such as imaging and genomics. These technologies are currently being applied to the most important questions in medicine; however, they require a large amount of data to accurately capture relevant features in a population. In this proposal, we aim to use these cutting-edge machine learning approaches to gain new insight into the biology of advanced metastatic prostate cancer and develop novel biomarkers to guide the clinical management of men suffering from prostate cancer. To this end, we will develop novel computational tools that allow us to robustly extract valuable information from histology slides that are routinely used by pathologists. These studies will unmask previously unrecognized morphological features of cancer cells that can inform us about the biological behavior of a tumor. Next, we will elucidate how well-known changes in the tumor DNA can influence the appearance of cancer cells, which will allow us to predict the driving molecular changes in a tumor by just assessing the appearance of cancer cells. The ultimate goal of this research is to develop a simple tool for doctors that combines the complex information from the analysis of the biopsy tissues into a simple and interpretable score. This score will provide valuable clinical information and will allow us to predict the outcome and response to therapies of men with advanced prostate cancer. Prostate cancer is a heterogeneous disease without a one-size-fits-all treatment. Our team is driven by the idea that novel technologies can greatly increase the information content and the clinical value of metastatic biopsies. We hope that the new approaches described in this proposal will lead to a more personalized treatment for men with prostate cancer and ultimately reduce suffering from this devastating disease.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Dec 28, 2022
- Source ID
- W81XWH2210279
Entities
People
- Stephanie Harmon
Organizations
- National Cancer Institute
- United States Army