Computer-Aided Diagnosis and Feature-Guided Data Reduction Systems in Mammography
Abstract
We have been conducting the pilot clinical study to evaluate the effects of CAD on radiologists' reading of screening mammograms this year. We have analyzed the results of about 1,300 cases. The CADView system detected 100% (18/18) of the lesions that were recommended for biopsy in both sites, all fine needle biopsy cases (5/5) at the GU site, and missed only one of the fine needle biopsy cases (4/5) at the UM site. The CAD system detected both malignant cases at the UM site, whereas causing 19 additional callbacks and 1 additional benign biopsy. The CAD system detected all three cancers at the GU site, including one additional cancer that was not originally called by the radiologist, and only caused 2 additional callbacks. The CAD system also detected 74% (34/46) of the short-term follow up cases at the two sites. Since the number of cases collected so far is still small, we have not performed statistical analysis on the data yet. We will continue to collect cases at the UM and GU sites in the coming year. Two observer performance studies have been conducted for the CAD-guided image compression project. It was found that the proposed method with adequate bit rate will fully preserve the quality of microcalcifications and suspected microcalcifications without sacrificing the edge sharpness and overall image quality at an area-equalized bit-rate of about 0.4 bit/pixel. The CAD-guided compression can therefore reduce the image transmission and storage requirements for digital mammograms by a factor of about 30 without causing observable degradation of image quality. It can be an effective image compression method for picture archiving and communication and facilitate the implementation of telemammography and digital mammography.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 2000
- Accession Number
- ADA388204
Entities
People
- Heang-Ping Chan
Organizations
- University of Michigan