Computerized Analysis of MR and Ultrasound Images of Breast Lesions

Abstract

Although general rules for the differentiation between benign and malignant mammographically identified breast lesions exist, considerable misclassification of lesions occurs with the current methods. The main goal of the proposed research is to develop noninvasive, computerized methods for analyzing ultrasound and MR (magnetic resonance) images of breast lesions to aid radiologists in their workup of suspect lesions. We currently have retrospectively collected over 400 ultrasound cases of mass lesions, all that had gone on to either biopsy or cyst aspiration. We extracted and calculated features related to lesion margin, shape, homogeneity (texture) and the nature of the posterior acoustic attenuation pattern. Linear discriminant analysis round-robin runs yielded A(sub z) values of 0.94 and 0.87 in the task of distinguishing between benign and malignant lesions in the entire database and the equivocal database, respectively. The 'equivocal' database contained lesions that had been proven to be benign or malignant by either cyst aspiration or biopsy. Dynamic MR data was analyzed with the computerized classification method, which includes temporal features of normalized speed and inhomogeneity of uptake, and spatial features of margin descriptors such as circularity and irregularity. Results indicate that classification based on temporal and spatial features combined can yield a positive predictive value of 94%, and has the potential to reduce the number of unnecessary biopsies by approximately 92%. We also have developed a new method for automatically extracting the lesion from the 3D image set of the breast. ROC analysis yielded A(sub z) values of 0.90 when the manual segmentation was used in the classification and 0.93 when automatic segmentation was included.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jul 01, 2001
Accession Number
ADB277494

Entities

People

  • Maryellen Lissak Giger

Organizations

  • University of Chicago

Tags

DTIC Thesaurus Topics

  • Artificial Intelligence
  • Breast Cancer
  • Carcinoma
  • Computer Programs
  • Computer Vision
  • Computer-Aided Diagnosis
  • Computers
  • Data Science
  • Databases
  • Detection
  • Electronic Mail
  • Health Services
  • Information Science
  • Machine Learning
  • Magnetic Resonance
  • Medical Personnel
  • Ultrasounds

Readers

  • Medical Imaging.
  • Oncology and Biomarker-Based Cancer Detection.