Prediction and Clinical Application of Central Lymph Node Metastasis in Papillary Thyroid Carcinoma Based on Multi-modal Ultrasound Feature Fusion: A Multi-center Study.
Authors
Affiliations (7)
Affiliations (7)
- Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, Hefei, China.
- School of Basic Medical Sciences, Anhui Medical University, Hefei, China.
- Hefei Cancer Hospital of CAS, Institute of Health and Medical Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, China.
- Information Materials and Intelligent Sensing Laboratory of Anhui Province, Anhui University, Hefei, China.
- Hefei Cancer Hospital of CAS, Institute of Health and Medical Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, China. Electronic address: [email protected].
- Hefei Cancer Hospital of CAS, Institute of Health and Medical Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Hefei, China. Electronic address: [email protected].
- Department of Ultrasound, The First Affiliated Hospital of Anhui Medical University, Hefei, China. Electronic address: [email protected].
Abstract
To develop and validate a deep learning model integrating multi-modal ultrasound information from B-mode ultrasound (BMUS) and strain elastography (SE) for accurate preoperative prediction of central lymph node metastasis (CLNM) in patients with papillary thyroid carcinoma (PTC). In this retrospective multi-center study, 568 patients with PTC from 4 hospitals were enrolled and divided into a training set (n = 400), an internal validation set (n = 100) and an external test set (n = 68). BMUS and SE images were collected for each patient. After image pre-processing, deep features were extracted from each modality using EfficientNet-B4 and fused for CLNM prediction. Model performance was evaluated using the area under the curve (AUC), accuracy, sensitivity and specificity. Calibration and decision curve analysis were further performed. Diagnostic performance was further compared with radiologists of different experience levels, with and without artificial intelligence assistance. In the internal validation set, the multi-modal fusion model achieved an AUC of 0.929, outperforming the BMUS-only (AUC = 0.843) and SE-only (AUC = 0.875) models. In the external test set, the model maintained good generalizability with an AUC of 0.843. Calibration and decision curve analysis demonstrated good agreement and clinical net benefit. Grad-CAM visualizations demonstrate a potential alignment between the regions attended to by the model and clinical signs. The proposed model outperformed all individual radiologists (AUC range, 0.585-0.649), and artificial intelligence assistance improved radiologists' diagnostic performance (AUC increased to 0.701-0.752) and confidence. The proposed multi-modal deep learning model enables accurate and interpretable preoperative prediction of CLNM in PTC patients, demonstrating robust performance across centers and effective clinical decision support potential.