GSO-CNN-based model for the identification and classification of thyroid nodule in medical USG images

Thyroid ultrasonography is one of the widely used techniques for the detection and classification of thyroid nodules. In this paper, grid search optimization (GSO)-based convolutional neural network (CNN), i.e., GSO-CNN model is proposed for thyroid nodule identification and classification. A total of 295 public and 654 collected thyroid USG datasets are considered in this work. The increased datasets size for the proposed model becomes 1770 for the public dataset and 3924 for the collected dataset after applying data augmentation techniques. We experimentally determined the best optimized value using grid search optimization (GSO) technique for learning rate and dropout. The model works in four phases: (i) data collection, (ii) pre-processing, (iii) morphological operation, segmentation and boundary detection and (iv) classification using CNN. The proposed model has achieved an accuracy of 95.30%, sensitivity of 96.66%, specificity of 94.87% and f-measure of 97.20% on the public dataset having 1770 thyroid USG images and an accuracy of 96.02%, sensitivity of 96.70%, specificity of 95% and f-measure of 98.34% on the collected dataset having 3924 thyroid USG images. The proposed model has been compared with popular deep learning techniques like dense neural network (DNN), Alexnet, Resnet-50 and Visual Geometry Group (VGG-16) with and without considering segmentation and boundary detection techniques. The proposed model has shown an improvement of (6.126%, 6.846%), (7.1%, 7.14%), (6.77%, 6.9%) and (7.77%, 8.91%) in terms of accuracy, sensitivity, specificity and f-measure on (dataset -1, dataset-2) against other state of the art models.