Algorithm mediated early detection of oral cancer from image analysis.

ABSTRACT

OBJECTIVE: To develop Automatic Oral Cancer Detection algorithm for identification and differentiation of premalignant lesions from buccal cavity images for early detection of oral cancer, which may reduce related fatalities in developing countries. STUDY DESIGN: The oral cavity images of normal, erythroplakia, and leukoplakia (20 images of each) were collected and processed using MATLAB image processing tools. First, maximum red value was used to differentiate between normal and abnormal. Second, mean red value was used for the selection of a processing path through YCbCr. Third, gray-level co-occurrence matrix (GLCM) based features were used to make final decisions. Images have been randomly divided and shuffled between training and test set to rigorously train the algorithm. RESULTS: With 100% efficiency, normal images were separated from abnormal images in the first step by applying R value distribution with a cutoff R value, 11,900. Further, images with a mean R value >200 and <200 were processed by segmentation of Y plane and Cr plane, respectively. For the final decision, abnormal images were analyzed through the GLCM using the entropy feature as one of the key indicators, which can apply to the differentiation decision with 89% efficiency. CONCLUSIONS: The developed algorithm can successfully differentiate premalignant lesions from normal. A graphic user interface was developed, which displays outcomes with reasonable accuracy.