Electrical Impedance-based Classification for In Vivo Oral Cancer Detection

In oral cancer surgery, surgeons struggle to delineate where the cancer ends and healthy tissue begins, posing a challenge to achieve negative margins and a complete resection. Intraoperative electrical impedance-based devices can help classify oral cancer and guide surgeons to establish where positive margins remain. New machine-learning algorithms can enhance classification performances and enable real-time results. This paper focuses on evaluating the performance of several machine learning algorithms to identify cancerous tissue samples from in vivo tissue measurements. Electrical impedance measurements of healthy, benign, dysplastic, and cancerous tissue were taken intraoperatively during oral cancer resection surgeries. Cancerous lesions were successfully identified with high accuracy in the binary classification setting reaching AUCs of 0.92. Algorithm classification performance of all tissue types was evaluated as a multiclass setting as well, achieving AUCs of 0.81. Future work involves optimizing models and exploring additional models.