Automatic Cellularity Assessment in Surgical Specimens After Neoadjuvant Therapy of Breast Cancer

The goal of this research is to develop methodologies to automatically assess residual tumor cellularity in surgical specimens after neoadjuvant chemotherapy for breast cancer. At present pathological examination of the tissue sections after surgery is the gold-standard to estimate the residual tumor. Currently, tumor cellularity is manually estimated by pathologists from hematoxylin and eosin stained slides. The quality, and reliability of estimates might be impaired by inter-observer variability. For the past decade, pathology labs have been transforming their work flow by converting their physical tissue specimen into digital format enabling sample images to be archived, stored, and processed digitally. Furthermore, efficient computer algorithms can be employed to make the most of the embedded information in digital tissue slides and to detect important knowledge from within them. In this thesis, image analysis and machine learning techniques have been used to identify regions that contain clinically relevant information and to automatically estimate residual cancer cellularity from within regions of interest and whole slide images of breast cancer specimen. Methods developed in this thesis have been applied on actual complete patient cases and have been compared with other state-of-the-art techniques. In particular, for assessing cellularity of residual tumor, intra-class agreement coefficients of 0.74 (95% CI of [0.70, 0.77]) and 0.75 (95% CI of [0.71, 0.79]) were found between pathologists and the automated method respectively, while the agreement coefficient between the participating pathologists was found to be 0.89 (95% CI of [0.70, 0.95])Ph.D