Feature Selection of Oral Cyst and Tumor Images Using Principal Component Analysis

Abstract

Tumor and cyst are two dangerous gum diseases commonly found in the mouth. However, unnoticed signs and symptoms in the early stages of them frequently lead to the late treatment of recovery. Earlier detection to them as a preventive care before becoming a chronic cancer is considered important leading to earlier diagnosis and treatment. Feature selection before detection and classification plays a vital role in order to maximize the classification accuracy. In this research, an implementation of principal component analysis (PCA) is proposed to overcome the high dimensionality of the dental panoramic images. This research is intended to offer a solution in selecting the most dominant and principal features to prevent the features weaken the accuracy. It has figured out that by using PCA, there are only four features that dominant among 33 features extracted. This means that only 12% of overall features significantly play a dominant role. Variance of these features affects the proportion contributed. Components that have a proportion of contribution greater than 1% are PC1, PC2, PC3, PC4, each of 86.44%, 9.74%, 2.59%, and 1,125%. The four dominant features which have been found are Feature 21, 22, 24, and 27 extracted by using GLRLM with SRE, LRE, RP, and HGRE respectively in other words, the 4 selected features represent 99.7% of the overall data variance representing 99.7% of the overall data variance.