Pre-CAD normal mammogram detection algorithm based on tissue type.

摘要

Breast cancer is the second leading cause of cancer-related deaths in women in the United States. X-ray mammograms are one of the most common techniques used by radiologists for breast cancer detection and diagnosis. Early detection and diagnosis is important, since may statistics have shown that detecting the cancer in its early stage will reduce the mortality rates by 30 -- 70%. Although most CAD systems were designed to help radiologists in their diagnosis by providing useful insight, the accuracy of CAD systems remains below the level that would lead to an improvement in the overall radiologists' performance. Two main problems appear to affect the decision of detecting and diagnosing breast cancer: the accuracy of the CAD systems used, and the radiologists' performance in reading and diagnosing mammograms. In this work we help to improve CAD system's performance as well as radiologists' performance by adding a preprocessing step to improve the sensitivity significantly. In this work we developed a pre-CAD system that is based on separating mammograms into two disjoint categories according to their tissue type (fatty, or dense). Unlike other CAD systems who aim to detect abnormal mammograms, we are using our pre-CAD system to detect normal mammograms instead of abnormal ones. The pre-CAD system will work as a "first look" that will screen-out normal mammograms, leaving the radiologists and other conventional CAD systems to focus on the suspicious cases. This will reduce the workload of radiologists and allow them to focus on the "hard to classify" cases. Moreover, our pre-CAD system design is based on the separation of mammograms into fatty or dense according to their radiologically-defined breast density. This will improve the classification accuracy since the classifier will focus on detecting normal mammograms within the same tissue type. A one-class Support Vector Machine classifier is used to detect normal mammogram in tissue-type separately. This helps improve the overall performance of radiologists if it is used as a complement to an existing CAD system. Gray-level co-occurrence matrix (GLCM) and Local Binary Pattern (LBP) features were extracted for each of dense and fatty mammograms. The results showed that the classifier performance is significantly improved when GLCM features are extracted for fatty tissues. On the other side, the classifier performance was significantly improved when LBP features are extracted for dense tissues. The sensitivity was significantly increased when dense and fatty mammograms were separated. In summary, different set of features suited different tissue densities. Future work could focus on designing a fully-automated pre-CAD system for normal mammogram classification.