Predicting Central Lymph Node Metastasis in Papillary Thyroid Carcinoma Using Multimodal Data

During thyroid cancer surgery, the debate continues regarding whether to perform routine prophylactic central lymph node dissection on the same side as the lesion. Central lymph node metastasis has been proven to be clinically significant in guiding treatment decisions. Therefore, accurately predicting central lymph node metastasis preoperatively is an significant step in the diagnosis and treatment of thyroid cancer. In clinical practice, ultrasound examination is commonly used to assess thyroid cancer. However, due to the low contrast and high noise of ultrasound images, the accuracy of ultrasound in identifying lymph node metastasis is limited. This study proposes a preoperative prediction of lymph node metastasis based on MFF-Resnet, which is based on Resnet50. This paper constructs multimodal data by combining deep learning with radiomics and clinical factors, and uses auxiliary branches to incorporate original information and low-level features of ultrasound images for prediction tasks. The AUC of MFF-Resnet reaches 0.8576, with an accuracy of 0.8018, and an F1 score of 0.8281.