Improving Cancer Foci Detection in Prostate Cancer Using Multiparametric MRI/MRS (GCC 1261)
The investigators' goal is to develop a non-selective and non-invasive procedure to identify aggressive tumors and simultaneously identify their exact location in Prostate cancer patients undergoing radical prostatectomy by combining multiparametric MRI and machine learning techniques. The combination of multi-parametric MRI and machine learning (validated using histopathology) can lead to increased sensitivity and specificity of cancer foci in the prostate, and help in isolating aggressive from indolent tumors. This increased sensitivity and specificity may eventually lead to: a) a reduction in the number of patients that undergo unnecessary treatment, and b) enhance current treatment options by enabling the use of focused therapies. The investigators will recruit 15 patients with prostate cancer that are currently scheduled to undergo radical prostatectomy into the study. All patients will obtain an advanced MRI study prior to the radical prostatectomy. MRI scans will include a) high-resolution volumetric images using T1 and T2-weighted imaging, b) vascular images using dynamic contrast enhanced (DCE) imaging, c) biophysical microstructure images using diffusion-weighted imaging, and d) biochemical images using MR spectroscopic imaging. Following radical prostatectomy, a pathologist will grade the prostatectomy specimens based on standard of care (Gleason grading system). Correlations will be generated between the parameters obtained from scans and from clinical assessments.
|Study Design:||Observational Model: Case-Only
Time Perspective: Prospective
|Official Title:||Improving Cancer Foci Detection in Prostate Cancer Using Multiparametric MRI/MRS and Machine Learning to Better Manage the Disease|
- Primary Objective: distinguishing high-grade tumors vs. low-grade tumors and normal prostate [ Time Frame: 16 months ] [ Designated as safety issue: No ]
Whether advanced MR imaging techniques can be used to train machine-learning techniques to distinguish high-grade tumors from low-grade tumors and normal prostate. The machine-learning techniques will be trained using histopathology data as the ground truth.
To achieve this we will obtain volumetric images of the various tissue attributes (listed below) and match them to histopathology:
- Vascular permeability (ktrans) using dynamic contrast enhanced MRI (DCE-MRI)
- Morphological changes captured using T2 and diffusion changes using diffusion weighted MRI (DW-MRI)
- Metabolic signatures of (choline+creatine)/citrate) or CC/C using magnetic resonance spectroscopic imaging (MRSI)
- Correlate in vivo imaging findings to ex vivo histopathology using deformable image registration
- Develop a multiclass support vector machine (SVM) using the set of multi-parametric images as input, and use it predict a score akin to the Gleason score.
- Secondary Objective: non-invasive and quantitative test to accurately identify the tumor grade and location. [ Time Frame: 16 months ] [ Designated as safety issue: No ]Advanced MR imaging techniques can be used alone in classifying tumor grade. Datasets collected will be partitioned into subsets that will be used for testing the machine learning techniques. For example: 90% of the data will be used for training and the remaining 10% will be used for testing. This process will be repeated over a combination of different subsets. Our hypothesis is that Machine Learning methods will assist in analyzing the differences between aggressive tumors, indolent tumors, and normal tissue. We further hypothesize this analysis will help in synthesizing an imaging-based "score" that can identify an aggressive tumor from indolent tumors and normal tissue in new cases after training. We believe using multi-parametric MRI combined with an advanced machine learning technique can improve the sensitivity and specificity of tumor foci detection. This will result in a non-invasive and quantitative test to accurately identify the tumor grade and location.
|Study Start Date:||April 2015|
|Estimated Study Completion Date:||December 2019|
|Estimated Primary Completion Date:||December 2017 (Final data collection date for primary outcome measure)|
Please refer to this study by its ClinicalTrials.gov identifier: NCT01766869
|United States, Maryland|
|Baltimore, Maryland, United States, 21201|
|Principal Investigator:||Warren D'Souza, PhD||UMD|