A CNN-SVM hybrid model for the classification of thyroid nodules in medical ultrasound images

Abstract

The thyroid nodule is one of the endocrine issues which is caused by the formation of irregular cells in the thyroid region. The recent success of machine and deep learning techniques in image recognition task leads to solve challenges in diagnostic imaging. An effective Convolutional Neural Network-Support Vector Machine (CNN-SVM) hybrid model is proposed using hinge loss function to achieve better results and stable convergence. The efficiency of the proposed model has been evaluated on public and collected data sets having 1180 and 2616 thyroid USG images after data augmentation. The proposed model has achieved an accuracy of 94.57%, specificity of 91.89%, sensitivity of 96.70% and f-measure of 95.64% on dataset-1 and an accuracy of 96%, specificity of 93.93%, sensitivity of 97.80% and f-measure of 98.33% on dataset-2. It has shown an improvement of 3% to 5% on dataset-1 and 4% to 6% on dataset-2 in comparison with state-of-the-art models.