Wavelet Representations for Digital Mammography

Abstract

We report on a receiver operating characteristics (ROC) study focusing on dyadic wavelets for enhancement of mammographic features in digitized mammograms. The enhancement protocol was based on multiscale expansions and non-linear enhancement functions described previously in our annual reports. In this study, dyadic spline wavelet functions were used together with a sigmoidal non-linear enhancement function. We designed a prototype test bed interface and performed a ROC study with three radiologists specialized in mammography. Data was obtained from the national mammography database of digitized radiographs from the University of South Florida. An initial analysis of the data counted the number of false-positives and true-positives in each of two cases: Enhanced and Original. Lesions with a LOC greater or equal 3 were considered malignant. The average TPF was found to be 0.667 with enhancement, and TPF = 0.569 without enhancement. This observed increase in sensitivity is encouraging, but accompanied by a slight increase in the fraction of false-positives (0.222 compared to 0.178). However, when the analysis of the data only focused on micro-calcifications alone, we observed a TPF 0.417 with enhancement compared to TPF = 0.222 without enhancement. No increase or decrease in FPF was observed. This finding reinforces our hypothesis that feature specific enhancement protocols are indeed useful for visualizing subtle mammographic features.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Apr 01, 1999
Accession Number
ADA381680

Entities

People

  • Andrew F. Laine
  • Fred Taylor

Organizations

  • University of Florida

Tags

Communities of Interest

  • Energy and Power Technologies
  • Human Systems

DTIC Thesaurus Topics

  • Computer Vision
  • Computers
  • Data Analysis
  • Databases
  • Detection
  • Graphical User Interface
  • Image Processing
  • Image Reconstruction
  • Information Science
  • Maximum Likelihood Estimation
  • Medical Personnel
  • Models
  • Operating Systems
  • Physicians
  • Signal Processing
  • Test Beds
  • Two Dimensional

Fields of Study

  • Medicine

Readers

  • Image Processing and Computer Vision.
  • Oncology and Biomarker-Based Cancer Detection.
  • Wave Propagation and Nonlinear Chaotic Dynamics.