Electrical Impedance-based Classification for In Vivo Oral Cancer Detection.

Area of Science:

• Biomedical Engineering
• Oncology
• Machine Learning

Background:

• Oral cancer surgery faces challenges in distinguishing cancerous from healthy tissue, impacting complete resection and negative margins.
• Intraoperative electrical impedance (IE) devices offer a potential solution for real-time tissue classification.
• Advancements in machine learning (ML) algorithms can enhance the accuracy and speed of IE-based cancer detection.

Purpose of the Study:

• To evaluate the performance of various ML algorithms in identifying oral cancerous tissue samples.
• To assess the efficacy of IE measurements combined with ML for intraoperative margin assessment.
• To compare binary and multiclass classification performance for different tissue types.

Main Methods:

• Intraoperative IE measurements were collected from healthy, benign, dysplastic, and cancerous oral tissues during surgery.
• Several ML algorithms were applied to analyze the IE data for tissue classification.
• Performance was evaluated using Area Under the Curve (AUC) metrics for both binary and multiclass settings.

Main Results:

• High accuracy was achieved in binary classification of cancerous lesions, with AUCs reaching 0.92.
• Multiclass classification performance across all tissue types yielded AUCs of 0.81.
• The study demonstrates the potential of ML-enhanced IE measurements for intraoperative oral cancer detection.

Conclusions:

• ML algorithms show significant promise in accurately identifying oral cancerous tissues using intraoperative IE measurements.
• This approach can assist surgeons in achieving negative margins and complete tumor resection.
• Further optimization and exploration of additional ML models are warranted for clinical application.