Explainable multimodal deep learning for predicting thyroid cancer lateral lymph node metastasis using ultrasound...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate preoperative prediction of lateral lymph node metastasis is vital for thyroid cancer management.
• Current prediction methods lack precision, impacting surgical strategies and patient prognosis.

Purpose of the Study:

• To develop and validate a deep-learning model for precise preoperative prediction of lateral lymph node metastasis.
• To integrate multimodal data for enhanced diagnostic accuracy.

Main Methods:

• Developed the Lateral Lymph Node Metastasis Network (LLNM-Net), a bidirectional-attention deep-learning model.
• Fused multimodal data from 29,615 patients and 9836 surgical cases across seven centers.
• Integrated nodule morphology, position, clinical text, and demographics.

Main Results:

• LLNM-Net achieved an Area Under the Curve (AUC) of 0.944 and 84.7% accuracy in multicenter testing.
• The model outperformed human experts (64.3% accuracy) and previous models by 7.4%.
• Tumors near the thyroid capsule (>72% metastasis risk) and specific lobes were identified as high-risk regions.

Conclusions:

• LLNM-Net demonstrates significant potential for preoperative screening and risk stratification of lateral lymph node metastasis.
• The model's ability to integrate diverse data sources enhances its predictive power.
• LLNM-Net offers a promising tool to improve surgical decision-making and patient outcomes.