Detection of benign and malignant thyroid nodules: an reinforcement region selection network framework

Thyroid nodules are the most common disease in the thyroid gland, where malignant nodules and some benign nodules can trigger thyroid cancer. Several computer-aided diagnostic methods have achieved certain results in identifying thyroid nodules. However, most of these methods focus on increasing the complexity of network structure or adjusting relevant parameters, while neglecting the role of the accuracy and completeness of texture feature information in nodule image in nodule discrimination. Therefore, this article proposes an RR EfficientNet network that utilizes reinforcement learning to improve the ability of nodule feature extraction. This network improves the accuracy of nodule location obtained by the Regional Proposal Network (RPN) through reinforcement learning, and trains the obtained complete image data of nodules in the EfficientNet network. Through relevant experimental analysis, it is shown that under the standard datasets, the accuracy of using this network model reaches 98.32%, the accuracy reaches 95.865%, and the recall rate reaches 93.84%. Its performance indicators are significantly improved compared to other methods.