Automated Framework for Effective Identification of Oral Cancer Using Improved Convolutional Neural Network

Abstract

Oral cancer is one of the most prevalent cancers worldwide, influencing and originating from the mouth and neck, and its prevalence is rising in many communities. Tobacco usage and smoking cigarettes are the leading causes of mouth cancer. A major issue continues to be the high incidence rate, delayed diagnosis, and inadequate treatment planning. Early diagnosis is crucial for a better prognosis, course of therapy, and chance of survival. Machine learning techniques have been hailed for improving diagnosis, which would subsequently reduce cancer-specific mortality and morbidity. In order to provide the effective identification of oral cancer, numerous machine learning algorithms has been used over the decade. Though machine learning algorithms are producing high level accuracy in classification task, but feature extraction is quite limited. Hence our current research utilised deep learning approach and developed the automated framework which encompasses of Pre-processing, feature extraction and classification. For effective pre-processing, contrast-limited adaptive histogram equalization (CLACHE) is used, which helps to intensify the contrast property over images and transforms the images into higher resolution. Then Gray Level Cooccurrence Matrix (GLCM) is used for feature extraction which helps to determine the features with respect to statistical texture. Finally Improved version of Convolutional Neural Network (ICNN) were employed for effective categorization of oral cancer. For analyzing the performance of developed framework, an oral cancer dataset has been used which was extracted from Kaggle Repository. Also, comparison has been made with various high performance existing approaches. As a result, our proposed framework achieves the accuracy of 97.32%, which was comparably better result than existing state-of-art approaches.

Topics & Concepts