Thyroid Nodule Identification in Ultrasound Images using CNN

Abstract

Early detection of any disease assists in reduction of disease progression. In this work discrimination of the thyroid nodules is studied by employing deep learning
architectures. Deep learning algorithm mimics the function of the human cerebral cortex. Convolutional Neural Network (CNN) is based on deep neural network design which has many hidden layers that are used to train a large dataset.

When the number of hidden layers is increased, it has a greater impact on the accuracy. In this study, CNN is employed for assessing the severity of the disease. In CNN, the output of the input layer is fed the convolutional layer followed by ReLU layer and Max pooling layer and the images are classified into Benign (TIRADS 2, TI-RADS 3, TI-RADS 4a) and Malignant (TI-RADS 4b, TI-RADS 5, TIRADS 6).

The performance of CNN is compared with the pre-trained networks such as Alex net, VGG-19 and Resnet-50 using transfer learning. The CNN outperformed the pretrained networks with an accuracy of 99.17%.