Development of Multiparametric MRI Model of Clinically Significant Prostate Cancer
Abstract
A reliable method to differentiate men with life-threatening (aggressive) versus non-life-threatening (indolent) prostate cancer does not currently exist. Without this critical information to help make treatment decisions, many men are over-treated when diagnosed with prostate cancer even when they have indolent disease. A reliable test to determine the aggressiveness of disease would be important for (1) determining the best treatment, (2) monitoring the cancer during treatment, (3) rapidly evaluating new drugs or therapies in clinical trials, and (4) determining early efficacy of "patient-tailored" treatments. If the tools were available to accurately monitor prostate cancer before and after treatment, great improvements in patient care and therapeutic developments could be made. Multi-parametric magnetic resonance imaging (mpMRI) is now a routinely used tool for the diagnosis and monitoring of prostate cancer, but currently, assessment of mpMRI is primarily qualitative. This requires radiologists to make a subjective assessment of the resulting images that is dependent on the physician s experience and training. An alternate approach is to take a quantitative approach that uses statistical models to combine multiple MRI parameters into a single prediction used to identify prostate cancer and, specifically, clinically significant disease. Quantitative approaches would be less subjective and would standardize the assessment of mpMRI independent of the physician s experience and training. The primary objective of this research is to develop an mpMRI classifier for clinically significant prostate cancer, defined as prostate cancer with Gleason score greater than or equal to 4+3. An mpMRI classifier for clinically significant prostate cancer would provide physicians with a non-invasive tool for diagnosing and evaluating the aggressiveness of prostate cancer. A non-invasive predictive model for clinically significant prostate cancer would allow clinicians to complete targeted biopsies that focus on regions of likely cancer, which would result in less missed cancers and more accurate grading. In addition, a non-invasive classifier for prostate cancer could be used to triage men with indolent disease to active surveillance, thus decreasing the number of men unnecessarily receiving treatment. To achieve this objective, we will improve and expand our correlative pathology dataset used to develop our statistical classifier and develop new statistical techniques for combining mpMRI into a predictive model for prostate cancer that allows us to assess cancer location, extent, and clinical significance. Improving our correlative pathology dataset will allow us to train our model on data that more accurately represents true disease status. This will result in an improved statistical classifier that more accurately discriminates between clinically significant and indolent disease. The novel statistical techniques developed in this proposal will allow us to make best use of the expanded correlative pathology dataset acquired in this application. mpMRI data have several unique features that we will exploit during model development. This will result in more accurate predictive models that fully utilize the unique dataset used to train our model. In addition, the development of new statistical methods will provide powerful tools for developing future classifiers and the development of computational tools to implement these models will allow us to easily update our model in the future if new or improved data become available. The development of a non-invasive mpMRI classifier for clinically significant prostate cancer has the potential to change clinical practice by providing a tool to facilitate a paradigm shift in treatment management towards active surveillance and away from immediate definitive therapy upon diagnosis. The methods developed would enable clinicians and patients to confidently initiate active surveil
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 04, 2016
- Source ID
- W81XWH1510478
Entities
People
- Joseph S Koopmeiners
Organizations
- United States Army
- University of Minnesota