An Effective Automated Framework for Oral Cancer Detection by Enhanced Convolutional Neural Networks

Oral cancer, an aggressive form of cancer affecting the mouth and throat, has been on an alarming worldwide increase due to tobacco use and smoking. Oral cancer presents various challenges such as high incidence rates, delayed diagnosis, inadequate treatment plans and late prognoses and outcomes; early detection is crucial in improving prognosis outcomes and survival prospects. Machine learning techniques have gained prominence for improving cancer diagnosis while decreasing mortality and morbidity; in recent years they were used to detect oral cancer, providing highly accurate classification tasks yet limited capabilities in terms of feature extraction capabilities. To address these limitations, we are currently employing a deep-learning approach. To this end, we have devised an automated framework encompassing feature extraction, preprocessing and classification. Contrast-limited adaptive Histogram Equalization (CLACHE), used for effective pre-processing purposes to increase image contrast and resolution, is one form of pre-processing while Gray Level Cooccurrence Matrix extracts statistical texture-based features; finally, a Modified Convolutional Neural Net (MCNN) was employed for accurate classification of oral cancer cases. Utilizing Kaggle Repository dataset, we used it to assess the performance of our framework, while comparing it against other high-performance methods. Our proposed framework had an accuracy rating of 99.13% which surpasses current state-of-the art approaches.