Deep Learning Method for Classifying Thyroid Nodules Using Ultrasound Images

The most prevalent endocrine malignant tumour is thyroid cancer, which is the fast growing cancer among all cancers. On palpation, nearly 5% of nodules in thyroid nodular diseases can be discovered, while 10% to 67% can be found on ultrasonography. Thyroid cancer progresses relatively slowly. The cause of thyroid cancer is inexpertly understood, but may involve genetic and environmental factors. If thyroid cancer is diagnosed at an early stage, it can be cured. Convolutional Neural Networks have a higher accuracy than standard Artificial Neural Networks in diagnosing thyroid nodules. For that reason, the proposed method uses the Convolutional Neural Network, a deep learning algorithm, in the classification of thyroid nodules using ultrasound images. The thyroid ultrasound image collection for this study was taken from the open access Thyroid Digital Image Database (TDID). The deep Convolutional Neural Network called Residual Network (ResNet) was employed as a state-of-the-art image classification model in this proposed method. The use of ResNet improves neural network performance. L2 regularization is introduced to prevent overfitting. The experimental results showed that the accuracy of using the ResNet model was 83%.