Combining Clinical, Sonographic, and Elastigraphic Features to Improve the Detection of Prostate Cancer
Abstract
The goal of this project is to combine features derived from ultrasound (US) images, US radio frequency (RF) data, tissue elasticity imaging, and clinical data such as PSA into a computerized system for displaying prostate images that indicate probable location(s) of cancer. This project began by gathering RF from in-vitro prostatectomy specimens in cross sectional planes 2mm apart. These data are used to calculate RF features such as backscatter coefficient at each location in the gland. The data are also used to generate images and elastograms from which image texture and tissue hardness features are computed. The features are correlated with histology from the same location in the prostate to determine which feature combinations accurately predict the presence of cancer. Despite the failure of Univ. of Texas to provide new elastography software and the abrupt departure of the laboratory assistant in June, significant progress in software development was made in the past year resulting in software that allows one to mark a region on a digital microscopic image of the prostate with automatic computation of RF and texture features for that region. Using this software, 55 benign and cancerous regions were studied. For this set of regions, RF slope and Co Occurrence Matrix Entcopy (ENT) were the best features for distinguishing cancer from benign tissue (Mahalanobis dist.= 1.498, Az = .77) . Some of the PP data acquired were acquired incorrectly necessitating acquisition of additional patients. Software to include elastographic features is being incorporated into the analysis package. With successful repair of the ultrasound scanner, acquisition of RF using an endorectal probe should begin within two months.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 01, 2001
- Accession Number
- ADA405506
Entities
People
- Brian Garra
Organizations
- University of Vermont