Automated Oral Cancer Detection using Deep Learning-based Technique

Oral cancer (OC) is a key worldwide health concern and is typically prevailing in low- and middle-income countries. Enabling the identification of patients suffering from potentially cancerous and noncancerous or normal lesions would lead to low-cost and early diagnosis of the disease and in turn reduce oral cancer mortality. During a routine oral examination, oral lesions are normally screened manually. Due to lack of medical facilities and medical expertise, the provision of an automatic initial screening process using artificial intelligence (AI) will greatly support to identify possibly cancerous lesions. The present work is an effort towards implementing an automated detection technique for diagnosing OC from oral photographic image dataset using deep learning techniques (DLTs). For this research, three candidate pretrained DL-CNN models specifically VGG19, InceptionNetV3, and Xception are preferred. These candidate DL-CNN models using the transfer learning (TL) approach were further improved with extra layers. The efficacy of the modified models is examined on an oral cancer photographic image database. It is examined that the pre-trained Xception DL-CNN model with modified structure has surpassed other models in terms of performance parameters.