Application of Convolutional Neural Networks in Thyroid Cancer Diagnosis and Analysis

Abstract

Due to the difficult and expensive diagnosis of thyroid cancer, this paper would like to discuss the possibility of using Convolutional Neural Network (CNN) model to assist doctors to classify and recognize medical images. Consequently, this article focuses on the three types of CNNs: Xception, VGG, and Inception. For the traditional models like VGG and basic Inception, they are trained for the basic image classification and recognition. Although they are only basic model in CNNs, they can also achieve a great success rate. For example, VGG-19 could deal with the four main types of thyroid cancer and the accuracy of the classification both over 98%. For the advanced model like Xception and InceptionResNet-v2, they are combined with various kinds of basic model and improved the parameters and structures. They are extensions and innovations based on the traditional models which also attained a successful result. For example, Xception could classify two kinds of medical images, the accuracy of ultrasound images is 0.98 and the accuracy of CT images is 0.96. According to analysis of the result, it is obviously found that the accuracy of each model could achieve a high rate. It proves that CNNs plays a significant role in diagnosis of the thyroid cancer. However, there are also some challenges and difficulties that need to be improved, including the selection of datasets and more comprehensive system coverage.