Ultrasound-based nomogram to predict the recurrence in papillary thyroid carcinoma using machine learning

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Papillary thyroid carcinoma (PTC) recurrence is common and can be fatal.
• Predicting PTC recurrence is crucial for patient management and survival.
• Current prediction methods may not fully capture recurrence risk.

Purpose of the Study:

• To develop a predictive model for PTC recurrence.
• To integrate clinicopathological features with ultrasound radiomics.
• To create a nomogram for enhanced recurrence prediction in PTC patients.

Main Methods:

• Extracted radiomics features from ultrasound images of 554 PTC patients.
• Utilized Cox regression and LASSO for feature selection.
• Constructed a combined nomogram using clinicopathological data and radiomics signatures.
• Validated the nomogram's performance using ROC curves, calibration curves, and DCA.

Main Results:

• The combined nomogram achieved AUCs of 0.851 (training) and 0.885 (validation).
• The nomogram demonstrated superior predictive performance compared to models using only clinical or radiomics data.
• Higher radiomics and risk scores were significantly associated with lower recurrence-free survival (RFS).

Conclusions:

• A nomogram integrating clinicopathological variables and ultrasound radiomics offers excellent performance for PTC recurrence prediction.
• The developed nomogram is a valuable tool for assessing recurrence risk in PTC patients.
• This approach enhances the ability to predict and manage PTC recurrence.