Thyroid Nodule Classification for Physician Decision Support Using Machine Learning-Evaluated Geometric and...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thyroid nodule classification relies on the Thyroid Imaging Reporting and Data System (TIRADS) guidelines.
• Current TIRADS classification involves subjective assessment of visual and textural nodule characteristics.
• Reducing subjectivity is crucial for accurate thyroid nodule diagnosis.

Purpose of the Study:

• To develop an objective method for classifying thyroid nodules using geometric and morphological (G-M) features.
• To reduce subjectivity in the current diagnostic process for thyroid nodules.
• To provide physicians with a decision support tool for thyroid nodule diagnosis.

Main Methods:

• Extracted 27 G-M features from ultrasound (US) thyroid nodule images.
• Selected 11 significant G-M features aligned with TIRADS criteria.
• Evaluated feature performance using machine learning (ML) for nodule classification.

Main Results:

• G-M features combined with ML achieved high accuracy, sensitivity, and specificity in thyroid nodule classification.
• The proposed method demonstrated superior performance compared to state-of-the-art approaches.
• Identified key G-M features that correlate with TIRADS visual characteristics.

Conclusions:

• Geometric and morphological features offer an objective approach to thyroid nodule classification.
• This method can serve as a computer-aided diagnostic (CAD) system, validating TIRADS criteria.
• The approach enhances diagnostic accuracy and supports clinical decision-making in thyroid imaging.