Automated Detection and Classification of Oral Precancerous Stages from White Light Images Using LightGBM

Abstract

Cancer is one of the major health challenges across the world and continues to cause a large number of deaths every year. Among different types, oral cancer is quite common and often goes unnoticed until it reaches an advanced stage. This delay in detection is a key reason for its high mortality rate. Identifying the disease at an early or pre-cancerous stage can greatly improve treatment outcomes and patient survival. In this work, a method is proposed to differentiate between non-cancerous and cancerous oral lesions while also identifying their early stages. The approach uses different color spaces to capture variations in images and extracts important color and texture features. These features are then analyzed using the LightGBM algorithm for classification. The results show strong performance across multiple evaluation measures, making the method both effective and efficient for practical use in early oral cancer detection.