Clinical Trial and Evaluation of a Computer Aided Diagnostic System for Breast Cancer Screening

Breast cancer is the second leading cause of cancer deaths in women, exceeded only by lung cancer. Earlier detection, through scanning, can help reduce the death rates from breast cancer. Mammography is the only non-invasive diagnosis method and it allows early detection of microcalcifications, circumscribed masses and speculated lesions. This increases the need for automated CAD systems that can accurately detect cancer cells at early stages. In screening mammography, radiologists face many challenges, which are:

  • Low contrast between the cancer cells and normal cells increases the difficulty of early detection.
  • The availability of quantum and structured noise.
  • Low conspicuity of the lesion.
  • Poor image quality (occasionally).
  • Eye fatigue, or oversight.
  • In this project we aim to develop an automated and reliable CAD system that can carry out two major tasks: detection of mammographic lesions, and diagnosis of cancer from identified lesions. However,the final decision regarding the likelihood of the presence of a cancer and overall patient management is left to the radiologist. This project is based on our existing work on mammography, where we have managed to develop a prototype of an automated and reliable CAD system that could detect microcalcifications in mammograms. This system is composed of three stages: pre-processing, segmentation and knowledge based microcalcification (MC) classification. These stages are designed to simulate the way radiologists detect MCs in mammograms and incorporate four visual characteristics of MCs. This prototype is tested on the two major benchmarks databases, which are the USF and MIAS databases and it achieves average detection rates of about 97% for true-positives.