Award Date

1-1-2007

Degree Type

Thesis

Degree Name

Master of Science (MS)

First Committee Member

Emma Regentova

Number of Pages

92

Abstract

Among signs of breast cancer encountered in digital mammograms radiologists point to microcalcification clusters (MCCs). Their detection is a challenging problem from both medical and image processing point of views. This work presents two concurrent methods for MCC detection, and studies their possible inclusion to a computer aided diagnostic prompting system. One considers Wavelet Domain Hidden Markov Tree (WHMT) for modeling microcalcification edges. The model is used for differentiation between MC and non-MC edges based on the weighted maximum likelihood (WML) values. The classification of objects is carried out using spatial filters. The second method employs SUSAN edge detector in the spatial domain for mammogram segmentation. Classification of objects as calcifications is carried out using another set of spatial filters and Feedforward Neural Network (NN). A same distance filter is employed in both methods to find true clusters. The analysis of two methods is performed on 54 image regions from the mammograms selected randomly from DDSM database, including benign and cancerous cases as well as cases which can be classified as "hard cases" from both radiologists and the computer perspectives. WHMT/WML is able to detect 98.15% true positive (TP) MCCs under 1.85% of false positives (FP), whereas the SUSAN/NN method achieves 94.44% of TP at the cost of 1.85% for FP. The comparison of these two methods suggests WHMT/WML for the computer aided diagnostic prompting. It also certifies the low false positive rates for both methods, meaning less biopsy tests per patient.

Keywords

Aided; Clusters; Computer; Detecting; Diagnostic; Digital; Inclusion; Mammograms; Microcalcification; Prompting; Study; Systems

Controlled Subject

Electrical engineering

File Format

pdf

File Size

2.73 MB

Degree Grantor

University of Nevada, Las Vegas

Language

English

Permissions

If you are the rightful copyright holder of this dissertation or thesis and wish to have the full text removed from Digital Scholarship@UNLV, please submit a request to digitalscholarship@unlv.edu and include clear identification of the work, preferably with URL.


Share

COinS