Performance Comparison of Oral Cancer Classification with Gaussian Mixture Measures and Multi Layer Perceptron

Abstract

One of the most commonly occurring cancers is oral cancer. The incidence of the oral cancer seems to be increasing exponentially in the world. The clinician has to undergo a higher level of dilemma every time in order to differentiate the cancerous lesions from other controversial and poorly defined lesions that are present in the oral cavity. Early stage carcinomas and its subsequent manifestations are highly misinterpreted because at the initial stage there is minimum discomfort in the patient and they simply mimic many similar benign lesions. The analysis to be done by the doctors is often delayed and therefore there is a high risk for the cancer to spread in the body. Squamous cell carcinoma is the most common malignant neoplasm present in the oral cavity. Therefore the accurate diagnosis and management of this particular Squamous cell carcinoma which originates from the surface of the oral muscle has to be done well. The main aim of this work is to assess the clinical features, diagnostic procedures and treatment required for oral cancer patients. The staging of the cancer is generally divided into two stages namely, clinical and pathological. In TNM (Tumour, Node, Metasis), a lot of novel prognostic tools have been traced and new methodologies for the prognostic factors have been drastically improved and developed. This paper compares the classification accuracy of the TNM staging system with the aid of Multi Layer Perceptron (MLP) and Gaussian Mixture Model (GMM) classifiers. In this work, totally 75 oral cancer patients are studied. For both the classifiers, the input variables are nothing but the TNM variables such as tumour size, number of positive regional nodes, distance metastasis, hereditary etc. Out of the two post classifiers utilized here, GMM provided a better result as of 94.18% average accuracy for all the stages while Multi Layer Perceptron (MLP) showed an average accuracy of about 89.5% for all the stages. In this paper, Extreme Learning Machines (ELM) is also employed as a post classifier later for the oral cancer classification and the performance of it is compared to the performance of both the GMM and MLP.