Multimodal Deep Learning Model Based on Ultrasound and Cytological Images Predicts Risk Stratification of cN0...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate preoperative risk stratification of cN0 papillary thyroid carcinoma (PTC) is crucial for effective treatment planning.
• Current methods may lack precision in assessing the risk of N0 PTC before surgery.

Purpose of the Study:

• To develop and validate a multimodal deep learning (DL) model for non-invasive, preoperative risk stratification of N0 PTC.
• To integrate preoperative ultrasound and cytological images for enhanced diagnostic accuracy.

Main Methods:

• A retrospective multicenter study involving 890 PTC patients for model development and validation.
• A deep learning model was trained and validated using ultrasound and cytological images.
• Model performance was assessed using metrics including AUC, accuracy, sensitivity, and specificity.

Main Results:

• The combined DL model achieved high performance, with an AUC of 0.922 in internal validation and 0.845 in the testing group.
• The DL model demonstrated superior diagnostic performance compared to existing clinical models.
• Image-based heatmaps provided interpretability for the model's risk stratification diagnosis.

Conclusions:

• A multimodal deep learning model integrating ultrasound and cytological data offers accurate risk stratification for N0 PTC.
• This AI-driven approach can significantly aid in guiding preoperative treatment decisions for PTC patients.