Project 466727

Neoadjuvant therapy (NAT) is a key stage in the treatment of early breast cancer. Being able to properly assess a patient's response to NAT provides important information on their survival and cancer recurrence rates. To standardize and quantify patient response, the residual cancer burden (RCB) index was created. To determine this index, pathologists must scan through a digital image of an entire microscope slide. However, because of the high magnification that can be achieved from the microscope, these digital images often reach over 100,000 by 100,000 pixels in size. This makes the determination of the values required for the RCB index subjective and time consuming for medical professionals.To alleviate this issue, this project proposes to apply machine learning (ML) algorithms to automate the process of determining two important pieces of the RCB index: the tumor bed and tumor cellularity. The tumor bed describes the total extent of the cancerous tissue, and the tumor cellularity describes the percentage of the tumor bed containing cancerous tissue. Previous research on this same topic has attempted to determine these values for an entire slide with limited success. This investigation will look to apply anew ML model called a Vision Transformer (ViT) to determine both of these values simultaneously, for a whole microscope slide. From previous experimentation, the ViT has shown excellent promise in summarizing information in images.The proper development of this research could lighten pathologist workflows, opening their time for other key tasks. This could have the effect of reducing patient wait times and decreasing the number of patients that are improperly recommended for further treatment, or lack thereof, by the RCB index.