Machine Learning-Assisted Diagnostic System for Indeterminate Thyroid Nodules.

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Indeterminate thyroid nodules (ITNs) require accurate diagnosis to guide treatment.
• Fine-needle aspiration (FNA) can yield inconclusive results for ITNs (Bethesda classes III, IV, V).
• Surgical excision is often performed for ITNs, carrying associated risks and costs.

Purpose of the Study:

• To develop and validate an ultrasound-based machine learning classifier for diagnosing benignity in ITNs.
• To assess the model's performance using surgical histopathology as the gold standard.
• To evaluate the potential of the model to reduce unnecessary surgeries for specific Bethesda classes.

Main Methods:

• A dataset of 194 ITNs from 180 patients was analyzed.
• Ultrasound features were scored using the American College of Radiology Thyroid Imaging Reporting and Data System (TI-RADS).
• A support vector machine (SVM) model was trained using eight parameters, including TI-RADS scores, nodule size, age, and sex.

Main Results:

• The SVM model achieved a sensitivity of 93.8% and a specificity of 56.5% in the testing set.
• The model demonstrated high negative predictive values (NPVs) for Bethesda III (93.9%) and IV (93.8%) categories.
• Histopathology confirmed 46.4% malignancy and 53.6% benignity in the analyzed ITNs.

Conclusions:

• An ultrasound-based SVM model shows promise in accurately classifying benign ITNs.
• The high NPV suggests potential to avoid surgery for selected Bethesda III and IV ITNs.
• This machine learning approach offers a pathway to optimize surgical decision-making for indeterminate thyroid nodules.