Classification of round lesions in dual-energy FFDM using a convolutional neural network. Simulation study
In this work with the FDA, we investigate if a convolutional neural network (CNN) can be trained to distinguish solid masses from cystic fluid in mammogram images. These round lesions account for over 20% of recalls of mammogram screenings. Masses are often malignant while cysts are almost always benign, so being able to discriminate between the two can alleviate stress on patients and save costs in the healthcare system. We simulate a full-field dual mammography (FFDM) system and use a realistic anthropomorphic digital breast phantom model to generate x-ray images. A ResNet CNN architecture classifies the regions of interest (ROIs) containing either cystic fluid or solid tissue.
Zoom: https://arizona.zoom.us/j/98593835992 Password: 021573