Identification of Calcified Polyps Through Texture Analysis

Abstract

A polyp is a small unhealthy growth on a surface inside one’s body. The texture of polyps can be classified as either hard or soft, depending on the calcification on it. The variations in textures result in several conditions that may disrupt the normalcy of a healthy body. The ultrasound scan detects the presence of a polyp, but it does not analyze the texture of the same. In this paper, the entire arrangement has been divided into five main phases. First, the acquired image undergoes segmentation. Next, the characteristics of the polyp are extracted by combining the concepts of Gabor Wavelets and Cumulative Distribution Function. Subsequently, a Convolutional Neural Network (CNN) model is created. The CNN takes the covariance matrix obtained in the previous phase as input and trains itself to implement the classification. Once the validation is performed, the validation accuracy and validation loss are obtained as the output. These metrics help to measure the improvement or deterioration in the performance of the network.