My research uses computer vision and machine learning to improve medical imaging, focusing on breast and CT imaging. There are three specific projects:
(1) We design deep learning models to diagnose breast cancer from mammograms. We perform single-shot lesion detection, multi-task segmentation/classification, and image synthesis. Our goal is to improve radiologist diagnostic performance and empower patients to make personalized treatment decisions. This work is funded by NIH, Dept of Defense, Cancer Research UK, and other agencies.
(2) We create virtual breast models that are based on actual patient data and thus contain highly realistic anatomy. We transform these virtual models into physical form using customized 3D printing technology. With NIH funding, we are translating this work to produce a new generation of realistic phantoms for CT. Such physical phantoms can be scanned on actual imaging devices, allowing us to assess image quality in new ways that are not only quantitative but also clinically relevant.
(3) We develop computer-aided triage tools to classify multiple diseases in chest-abdomen-pelvis CT scans. We are building hospital-scale data sets with hundreds of thousands of patients. This work includes natural language processing to analyze radiology reports as well as deep learning models for organ segmentation and disease classification.