The goal is to improve the accuracy of breast screening programs and, most importantly, detect every suspicious case. We have developed a classifier that sensitively detects abnormalities in digital mammograms. These abnormalities include indicators of pathology such as calcifications, masses, tissue deformations and foreign matter. Generally, the classifier exhibits 100% sensitivity for abnormalities.
The classifier uses features calculated from mathematically transformed mammograms to identify an image as normal or suspicious. The classifier performed well on a public breast-image archive. Here we will optimize it using images from the NL Breast Screening Program. Once the technical optimization and final configuration has been determined, the classifier will be deployed as a second reader. Images deemed negative by the radiologist will be examined by the classifier, when suspicious images are detected they will be sent back to the radiologist for reexamination. If successful, this software might eliminate missed disease.