A model for predicting lymph node metastasis of thyroid carcinoma: a multimodality convolutional neural network...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate preoperative evaluation of cervical lymph node metastasis (LNM) in papillary thyroid carcinoma (PTC) is crucial for surgical planning.
• Current ultrasound imaging methods have limitations in predicting LNM status for PTC.
• This study addresses the need for improved LNM prediction accuracy in PTC.

Purpose of the Study:

• To evaluate the role of convolutional neural networks (CNNs) in predicting LNM for PTC.
• To assess the effectiveness of multimodality ultrasound data in CNN-based LNM prediction.
• To enhance the accuracy of preoperative LNM assessment in PTC patients.

Main Methods:

• Developed and evaluated CNN algorithms using data from 308 PTC patients with confirmed LNM status.
• Utilized a training set (80%) and a test set (20%) for model development and validation.
• Employed Residual Network 50 (Resnet50) for feature extraction from B-mode and contrast-enhanced ultrasound (CEUS) images, combining them for multimodality prediction.

Main Results:

• The triple-modality method (B-mode, CEUS, and ultrasound examination) achieved the highest accuracy (80.65%, AUC =0.831).
• This outperformed B-mode alone (69.00% accuracy) and a dual-modality approach (75.81% accuracy).
• Heatmaps visualized model attention areas, aiding in understanding model behavior and limitations.

Conclusions:

• A PTC lymph node prediction model using triple-modality features significantly improved prediction performance.
• The deep learning model effectively mimics human expert workflows and utilizes multimodal data.
• This approach offers valuable support for clinical decision-making in PTC management.