Intraoperative Histological Analysis of Squamous Cell Carcinoma Tumor Margins using a Convolutional Neural Network

Overview

Squamous Cell Carcinoma (SCC) is the second most common nonmelanoma skin cancer. If left untreated, SCC can be aggressive, with potential to deform adjacent tissue or metastasize. Most cases of SCC are treated using Mohs Micrographic Surgery (MMS), a surgical method for the excision of cancerous tissue and the rim of the surrounding normal tissue. During MMS, intraoperative histological analysis of tumor margins is performed to ensure complete tumor removal. However, this process is demanding due to the examination of multiple margins in contingent timeframes and challenges with frozen section specimen quality. Automated approaches to performing intraoperative margin assessment should be considered. This project presents a convolutional neural network (CNN) for the automated analysis of histology images of squamous cell carcinoma tumor margins. A dataset of 95 WSIs was collected from patients undergoing SCC tumor excision in the MMS setting. A model was constructed and trained with these datasets. Upon evaluation, this model demonstrates the ability to accurately and rapidly analyze histology images of tumor margins of SCC, with an AUC-ROC score of 0.923 and an average prediction time of 33 seconds per WSI. Integration of this algorithm in workflows could significantly assist pathological analysis of tumor margins during the surgical procedure.