Machine Learning-Based Prediction of Lymph Node Metastasis and Volume Using Preoperative Ultrasound Features in...

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Papillary thyroid carcinoma (PTC) frequently involves cervical lymph node metastasis.
• Accurate preoperative assessment of metastasis extent is crucial for surgical planning.
• Current methods may not fully capture the nuances of lymph node involvement.

Purpose of the Study:

• To develop and validate machine learning models for predicting cervical lymph node metastasis and metastatic volume in PTC.
• To integrate preoperative ultrasound characteristics into predictive algorithms.
• To enhance the precision of surgical planning for PTC patients.

Main Methods:

• Retrospective analysis of 573 PTC patients.
• Feature selection using univariate and Logistic Regression (LR) analysis (p < 0.05).
• Development of predictive models using K-Nearest Neighbors (KNN), Gradient Boosting Machine (XGBoost), and Support Vector Machine (SVM) algorithms.
• Performance evaluation using ROC curve, sensitivity, specificity, and accuracy on a validation cohort.

Main Results:

• The Gradient Boosting model showed the best performance for predicting lymph node metastasis (AUC: 0.784, accuracy: 73.8%).
• The Gradient Boosting model also excelled in predicting metastatic volume (AUC: 0.779, accuracy: 74.4%).
• The study included 320 patients with lymph node metastasis, categorized by volume.

Conclusions:

• Machine learning models integrating preoperative ultrasound features effectively predict lymph node metastasis risk in PTC.
• These models can optimize surgical planning, guiding lymph node dissection extent and personalizing treatment.
• This data-driven approach offers a paradigm for preoperative risk assessment in thyroid oncology.