Study on differentiating benign and malignant thyroid nodules based on CT multi-phase artificial intelligence models.

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Rising global incidence of thyroid nodules requires better non-invasive diagnostic methods.
• Current research on AI for thyroid nodule differentiation using multiphase CT is limited.

Purpose of the Study:

• To develop and evaluate AI models for differentiating benign from malignant thyroid nodules using multiphase CT imaging.
• To compare the performance of machine learning and deep learning models with clinical data.

Main Methods:

• Retrospective analysis of multiphase CT data from 604 patients with thyroid nodules.
• Development of machine learning and deep learning models using radiomics, DICOM data, and clinical factors.
• Performance evaluation using ROC analysis and Grad-CAM visualization.

Main Results:

• AI models incorporating imaging data significantly outperformed clinical-only models (AUC=0.811).
• Nomograms combining radiomics score (Rad-Score) or AI score (AI-Score) with clinical data achieved the highest diagnostic accuracy (AUC=0.885 and 0.881, respectively).
• Integrated AI and clinical data approaches were superior to single data type models.

Conclusions:

• AI models integrating multiphase CT radiomics or deep learning features with clinical data offer a robust method for thyroid nodule differentiation.
• These integrated models show potential for enhancing clinical decision-making in thyroid nodule diagnosis.