Classification of Oral Cancer Into Pre-Cancerous Stages From White Light Images Using LightGBM Algorithm

Abstract

Cancer is one of the foremost reasons for death worldwide, with nearly 10 million deaths noted in 2020. Globally, oral cancer ranks sixth when compared to other cancers. It is lethal because most cases are noticed at advanced stages, which can be prevented if screened for or treated early in the pre-cancerous stages, successively leading to a significant decrease in the mortality rate. In this work, a method is proposed that can effectively differentiate between benign and malignant oral cavity lesions and also classify their pre-cancerous stages. The method involves exploring five distinct color spaces and extracting color and texture features, which are then classified using a machine learning technique called Light Gradient Boosting Machine (LightGBM). The overall performance is promising, outperforming the state-of-art methods for the task of oral cancer classification, with a testing accuracy of 99.25%, precision of 99.18%, recall of 99.31%, f1-score of 99.24% and specificity of 99.31% for the binary classification, and testing accuracy of 98.88%, precision of 98.86%, recall of 97.92%, f1-score of 98.38% and specificity of 99.03% for multi-class classification. The proposed method used hand-crafted features and a machine-learning classifier, which uses limited resources and is less time-consuming.

Keywords

• Binary and multi-class classification
• color spaces
• early detection
• feature importance
• oral cancer
• white light images