Automated Thyroid Nodule Detection in Ultrasound Image using Optimal Neural Network Classifier

Wrushali M. Mendre; Ranjana D. Raut; P. T. Karule

doi:10.37591/joarb.v2i2.468

Abstract

Malignant thyroid nodule detection is an important issue in thyroid ultrasonography image processing. Literature reports many algorithms in this domain. This research aims to design optimal decision support system (DSS) for ultrasound thyroid nodule image characterization into benign and malignant classes. Thyroid nodule region is segmented to extract texture features from thyroid nodule ultrasound images acquired from 38 patients using gray level co-occurrence matrix (GLCM) and statistical properties. These features are selected as input parameters for neural networks such as radial basis function, generalized feed forward (GFF), multilayer perceptron (MLP), jordan and self organizing feature map (SOFM). Experimental results demonstrated that the SOFM gives 100% classification accuracy for both output classes that is normal thyroid and thyroid cancer with minimum mean square error of 0.00064423 for normal thyroid and 0.000529768 for thyroid cancer. Correlation factor also approaches towards unity.Keywords: DSS, RBF, GFF, MLP, SOFM, PCA, ANN Cite this ArticleMendre Wrushali M, Raut Ranjana D, Karule PT. Automated Thyroid Nodule Detection in Ultrasound Image using Optimal Neural Network Classifier. Journal of Advancements in Robotics. 2015; 2(2): 15–25p.