Analysis of oral squamous cell carcinoma into various stages using pre-trained convolutional neural networks

Abstract

Oral cancer is a commonly prevalent disease in the world. Cancer begins when alterations in healthy cells take place and grow gradually. Finally, it forms a mass called cancer. In general cancer can be classified into malignant and benign. In malignant, the cells can grow and multiply affects other parts of the body but, benign does not spread. Among the cancers in the South East Asia, the cancers in oral cavity ranks among the third most common types of cancer. Oral Squamous Cell Carcinoma is a highly prevalent oral cancer affecting the head and neck more than 90% than other parts of the body. Until now, classification of Oral Squamous Cell Carcinoma classification into various stages is based on the cytological and architectural change which relies on the pathologist. Every pathologist while assessing the lesions of Oral Squamous Cell Carcinoma into various stages led to mistakes. To overcome this, Computer Aided Diagnosis gives the exact stages of Oral Squamous Cell Carcinoma into poorly differentiated, moderately differentiated and well differentiated. In this work, two Convolutional Neural Network Architectures Inception-v3 and Resnet50 are used as feature extractors. Then, the derived features are given to Multi-class Support Vector Machine and Random Forest. Random Forest and Multi-class Support Vector Machine classifies the Oral Squamous Cell Carcinoma into various stages namely poorly differentiated, moderately differentiated and well differentiated. The features obtained from Resnet50 when given to Random Forest gives the satisfactory performance of 92.08 %.