Predicting Central Cervical Lymph Node Metastasis of Papillary Thyroid Carcinomas Using Multi-view Ultrasound Images

Thyroid nodule classification in ultrasound images is an important task for central cervical lymph node metastasis (CLNM) of papillary thyroid carcinomas (PTC). In clinical practice, nodules are usually evaluated using thyroid ultrasound from both horizontal and vertical perspectives. Due to the low contrast, high noise, and individual differences of ultrasound images, it has become a challenging problem. To address these, we propose a method to assist in diagnosing CLNM of PTC using multi-view ultrasound images and patient information. Our network consists of two modules, ROI extraction and multi-view classification network. First, we employ the popular semantic segmentation network, U \(^{2}\) -Net, on our clinical dataset and use the results to identify the region of interest (ROI). Then we design the parallel ResNet-50 network to complete the classification task from multi-view information. Experimental results show that our method can provide useful information for clinical diagnosis and lay a technical foundation for the classification of ultrasound images.