Texture Analysis of Thyroid Ultrasonography Images for Diagnosis of Benign & Malignant Nodule Using Feed Forward Neural Network

Abstract

A texture analysis of medical images gives the quantitative information about the tissue characterization & internal
structure of organs for possible pathology. The physician deducing the useful information concerning internal body parts for the
pathology or lesions. This is the till subjective matter of concern & thus to provide the objectification of disease diagnosis from the
medical image, this paper gives the idea for thyroid nodule diagnosis using texture of the ultrasound images. Thyroid gland is
located at the base of the neck, just below Adam’s apple which produces hormones that control body metabolism. The nodules are
found in thyroid may be benign or malignant. In this paper, gray level co-occurrence matrix (GLCM) is used as the texture
characterization technique. The 10 GLCM feature are selected for feature extraction & GLCM matrix is calculated for four
different orientation & different pixel distance from 1 to 15. The extracted features are classified using feed forward network
using Levenberg-Marquardt backpropagation optimized training algorithm for diagnosis of thyroid nodule malignancy risk. The
experimental results show the performance measure of feed forward neural network in terms classification accuracy, Positive
predicted rate, Negative predicted rate, sensitivity & specificity.