Correlated Preference based Weighted Feature Vector using ResNet50 for Accurate Oral Cancer Stage Detection

Abstract

With 130,000 deaths annually, Oral Cancer (OC) is the eighth most common cancer worldwide in India. The tumors that make up oral cancer can be found in many different places, including the salivary glands, tonsils, neck, face, and mouth. A biopsy, which involves taking a small piece of tissue from one area of the body and analyzing it under a microscope, is one of several diagnostic methods used to diagnose oral cancer. It is possible that the patient's oral glands, face, neck, or mouth will be impacted. The use of histological images in cancer screening aids in the detection and prognosis of abnormalities. Early diagnosis of oral potentially malignant disorders is crucial for improving the morbidity and mortality outcomes from mouth cancer. Early detection is often the key to a successful cure for many conditions, which have a chance of progressing to cancer. The potential applications of computer vision and deep learning algorithms in the detection of oral cancer using photographic images are analyzed to identify oral disorders that could be cancerous. Deep learning is applied in this research that recognized a wide range of oral and mouth diseases, including gum disease, canker sores, cold sores, oral lichen planus, oral thrush, mouth cancer, and oral cancer. When features are used incorrectly or excessively, classification algorithms learn a lot of useless information from images, leading to poor classification accuracy. A correlated preference based weighted feature vector is proposed to extract various types of properties from histopathology images. The texture and deep features obtained from these techniques are used as input vectors by the deep learning model. This research makes use of ResNet-50 architecture that makes use of 50 stackable bottleneck residual pieces. Traditional convolutional and pooling layers preprocess the image in the network's initial layers before the remaining blocks do any more processing. This research proposes a Correlated Preference based Weighted Feature Vector using ResNet50 for accurate Stage Detection (CPbWFV-SD) of oral cancer. The weighted feature vector is used to train the model and the minute change in the feature attribute set is used to identify the stage of the disease. The proposed model when compared with the traditional methods performs better in oral cancer stage detection.