Ultrasonic Classification of Multicategory Thyroid Nodules Based on Logistic Regression

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Thyroid nodules are common, and accurate diagnosis is crucial.
• Distinguishing benign from malignant nodules requires reliable predictive tools.

Purpose of the Study:

• To develop and compare logistic regression models for predicting thyroid nodule malignancy.
• To identify significant ultrasound features for nodule diagnosis.

Main Methods:

• Retrospective analysis of 1906 thyroid nodules from 1761 patients.
• Nodules categorized by size and inclusion of vascular/elastographic features.
• Univariate and multivariate analysis to screen significant sonographic features.
• Development and validation of logistic regression models.

Main Results:

• Hypoechoic, irregular margin, and microcalcification were significant features across all models.
• Predominantly solid and taller-than-wide shape were key for macronodules and micronodules, respectively.
• Strain elastography showed diagnostic value; models achieved AUC > 0.7.

Conclusions:

• Hypoechoic, irregular margin, and microcalcification are key indicators for malignant thyroid nodules.
• Developed models are effective for various nodule types and features.
• New models offer valuable diagnostic support for thyroid nodules.