Diagnostic Value of Machine Learning-Based Quantitative Texture Analysis in Differentiating Benign and Malignant...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Distinguishing benign from malignant thyroid nodules is crucial for appropriate patient management.
• Current diagnostic methods can have limitations, necessitating improved accuracy.

Purpose of the Study:

• To assess the diagnostic performance of machine learning (ML) quantitative texture analysis for thyroid nodule classification.
• To evaluate the ability of ML texture features to differentiate benign from malignant thyroid lesions.

Main Methods:

• Quantitative texture features were extracted from 235 thyroid nodules (102 malignant, 133 benign).
• A random forest ML classifier was employed for nodule differentiation.
• Feature selection and dimension reduction were performed using reproducibility testing and a wrapper method.

Main Results:

• 284 out of 306 (92.2%) texture features demonstrated good reproducibility.
• The ML model achieved a diagnostic accuracy of 86.8%, with 85.2% sensitivity and 87.9% specificity.
• The area under the curve (AUC) for the classification model was 0.92.

Conclusions:

• Quantitative texture analysis combined with ML classification shows significant potential for accurately discriminating benign and malignant thyroid nodules.
• Further validation through multicenter prospective studies with independent external data is recommended.