UC San Diego

Approximately a third of all breast conversation therapy patients require follow-up procedures due to positive margins in their final pathological diagnosis. While manual intraoperative evaluation of margin status can greatly reduce the need for secondary procedures, it requires detailed analysis by a highly skilled pathologist inside the operating room for accurate results. This dissertation documents the development of an automated approach to detect cancerous epithelial cells and evaluate fluorescently stained imprint cytology slides taken from excised breast tissue. By digitizing whole slide images and developing custom software algorithms to distinguish epithelial cells from debris and imaging artifacts, the presence of invasive cancer was detected with 95% accuracy with no false positives when validated against the diagnosis of a highly experienced pathologist. While the difference between a cluster of cells and an air bubble or fiber is readily apparent to the human eye, a digitized microscopy system required training to recognize subjects of interest in slide images. Performing statistical measurements of the fluorescent properties of objects revealed intrinsic differences between cells and debris that could not be appreciated when focusing on intensity alone. Using these measurements to train an algorithm to correctly identify epithelial cells was crucial to the overall success of the technique. Accurately identifying fluorescently stained cells from a population of objects in a digitized image could be readily adapted to other clinical specimens beyond imprint cytology slides of breast tumor margins. This work demonstrates a proof of concept for developing a highly accurate and automated digital pathology system for intraoperatively evaluating margin status to guide surgical decisions, reduce pathologist workload, and lower positive margin rates. Lastly, an overview of a novel platform for analyzing single cell secretants through fluorescent imaging and development of a microfabrication process to produce it will be detailed