Classifier Feature Fusion Using Deep Learning Model for Non-Invasive Detection of Oral Cancer from Hyperspectral Image

Abstract

In this paper we present a state-of-the-art infrastructure approach to detect and classify oral cancer from the hyperspectral imaging of investigating maxillofacial region. Oral and neck cancer is one of the rampant forms of cancer and this cancer is mostly experienced by socio-economic backward population. The hyperspectral image analysis is emerging as a non-invasive method for classification of cancer. Due to dearth of modern digital tool for computer-aided classification and pre-detection of cancer cells, we have proposed a Deep Boltzmann Machine (DBM) and SVM classification fusion for learning and classifying the pre- and post-cancerous tissue and normal tissue from the hyperspectral imaging. The mixed pixel from background is projected for cancerous region detection. The result of a patient hypercube is presented for the validation of deep learning technique pixel-wise probability map of cancerous and normal healthy tissues on hyperspectral imaging. Moreover, we have obtained a classifier accuracy of 94.75% by classifier fusion by majority voting as compared to conventional classification using the deep learning method imaging technique in hyperspectral image. Hence, the proposed digital pre-screening framework using deep learning classifier fusion on hyperspectral thermal imaging provides a high potential cancer identification tool for socio-economic backward patients in modern healthcare system.