Texture Analysis of Color Oral Images for Lesion Detection

Among all types of cancer, one of the prevalent forms is Oral Cancer. Diagnosis of oral cancer at an early stage and treatment of oral premalignant lesions reduces the mortality rate. Texture features are useful in identification of oral lesions as they have distinguishable patterns which cannot be easily recognized by a human observer. The objective of this work is to combine texture features and fractal features in analyzing color oral lesion images to identify malignancies. The proposed method was experimented on our own dataset of 200 digital color oral lesion images. The dataset was created by collecting images from different medical colleges and hospitals in Karnataka. A Back Propagation Neural Network has been trained using these features. The classifier detects abnormalities with an accuracy of 95%. Results indicate that the combined features have better potential in identifying benign and malignant oral lesions.