Deep learning predicts cervical lymph node metastasis in clinically node-negative papillary thyroid carcinoma.

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate staging of early-stage thyroid cancer is crucial for treatment planning.
• Cervical lymph node metastasis (CLNM) assessment is vital, but challenging in clinically node-negative patients.
• Current imaging methods often struggle to detect subtle lymph node involvement.

Purpose of the Study:

• To develop and validate a deep learning model for non-invasive prediction of CLNM in clinically node-negative papillary thyroid cancer (PTC).
• To integrate conventional ultrasound (US) imaging and clinical variables for enhanced predictive performance.
• To compare the model's diagnostic accuracy against expert radiologists.

Main Methods:

• An ensemble deep convolutional neural network (DCNN) model based on ResNet-50 was constructed.
• The model integrated brightness mode ultrasound (BMUS), color Doppler flow imaging (CDFI), and clinical data.
• A dataset of 1031 clinically node-negative PTC patients from two institutions was used for training, validation, and testing.

Main Results:

• The DCNN model demonstrated high predictive performance with AUCs of 0.86 (internal) and 0.77 (external) for CLNM.
• The model outperformed the average radiologist in accuracy (0.72 vs 0.59), sensitivity (0.75 vs 0.58), and specificity (0.69 vs 0.60) on the external validation set.
• The ensemble DCNN achieved superior test performance compared to expert interpretations.

Conclusions:

• Deep learning offers a non-invasive and accurate method for predicting CLNM in clinically node-negative PTC.
• Integrating US imaging AI with clinical variables enhances CLNM prediction.
• The developed DCNN model shows potential for improving diagnostic accuracy and guiding treatment strategies in thyroid cancer management.