Automated image processing algorithms to identify cancer within optical coherence tomography images of breast specimens
Early detection holds the key to successful treatment of breast cancer and leads to more flexible treatment options, including breast conserving surgery and non-surgery approaches. Our contribution towards this large goal is the development of automated algorithms to assess OCT images of lumpectomy specimens. The aims of this project were supported by the Columbia University Research Initiatives in Science and Engineering and the Herbert Irving Comprehensive Cancer Institute.
To enable future translation of this technology, automated analysis is needed to probe the datasets in a fast, non-subjective manner and localize suspicious areas of interest to give the surgeon direct feedback to confirm if the margins are clear. Deep learning will allow for an unbiased extraction of features, with the goal of improving our classification accuracy to be suitable for intraoperative guidance.