Comprehensive Summary
In this article, a new deep learning (DL) model for oral cancer (OC) diagnosis and classification is described. To develop this model, color normalization of histopathology images followed by tiling and augmentation were first performed. Following this, the key features were selected using the Weighted Fisher Score (WFS) to address any class imbalances. Using such feature-based inputs as opposed to full image uploads, the U-Net classifier employed has developed improved efficiency. Combining the aforementioned steps and techniques with Explainable Artificial Intelligence (XAI), this DL model uses multiple approaches to improve the classification of OC, addressing several current challenges in identifying OC. In particular, this DL model has a classification accuracy of 99.54%, exceeding the level of accuracy, precision, and reliability of all existing methods.
Outcomes and Implications
Oral cancer (OC) detection is critical to diagnose and treat OCs before they metastasize to other parts of the body. Currently, many cases of OCs are undetected, resulting in progression of the disease to a higher severity before detection and intervention. The deep learning (DL) model proposed in this article provides a novel method to detect OC presence, using several levels of accurate measures to contribute to the diagnosis. Additionally, the accuracy of this model exceeds all current methods. This DL model could be used as a tool alongside traditional diagnostic methods by health-care professionals to best assess the presence of OC in a patient. Because of the DL model’s reliability and precision, it may help providers detect OC early on and allow for medical intervention before the severity of the OC increases. In doing so, it may help improve patient health-outcomes and reduce the cost-burden on both the patient and the health-care provider.