Breast Ultrasound: Computer-Aided Diagnosis Approach to Improving Specificity and Decreasing Observer Variability

Abstract

The purpose of this grant is to construct an artificial neural network (ANN) to assist radiologists in differentiating benign from malignant solid breast lesions. The ultrasound (US) examinations and mammograms of sixty-four solid breast lesions that subsequently underwent histologic confirmation were evaluated in a blinded manner. Descriptive terms were chosen to characterize the ultrasonographic and mammographic appearance of the lesions from a pre-defined lexicon. In addition, patient age was recorded. These descriptive terms and patient age were used as inputs to train an artificial neural network to differentiate benign from malignant breast ma%ses. An ANN using only seven US descriptive terms as inputs performed better than a similar ANN previously constructed using mammographic descriptive terms and patient medical history as inputs. In addition, neural networks using combinations of inputs including US and mammogram descriptors and patient age all performed well for training an ANN. Further work includes constructing similar ANNs using more training cases, as well as testing these ANNs using prospective data. Significant interobserver variability in radiologists' descriptions and assessment of breast US exams was also demonstrated. The ANN may help decrease the variability in lesion assessment.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 1998
Accession Number
ADA344495

Entities

People

  • Jay A. Baker

Organizations

  • Duke University Hospital

Tags

DTIC Thesaurus Topics

  • Accuracy
  • Biomedical Research
  • Breast Cancer
  • Computer-Aided Diagnosis
  • Computers
  • Data Acquisition
  • Data Science
  • Data Sets
  • Databases
  • Medical Personnel
  • Neural Networks
  • North America
  • Observers
  • Physicians
  • Statistical Analysis
  • Training
  • Ultrasounds

Fields of Study

  • Medicine

Readers

  • Medical Imaging.
  • Neural Network Machine Learning.
  • Women's Health and Cancer Risk Research: African American Women and Pregnancy Outcomes.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks