A bayesian network for differentiating benign from malignant thyroid nodules using sonographic and demographic...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Thyroid nodules are common, with malignancy assessment crucial for patient management.
• Accurate differentiation between benign and malignant thyroid nodules remains a clinical challenge.

Purpose of the Study:

• To develop a Bayesian network (BN) for predicting thyroid nodule malignancy.
• To integrate diverse imaging features and patient demographics into a predictive model.
• To assist radiologists in evaluating the probability of malignancy in suspicious thyroid nodules.

Main Methods:

• A Bayesian network (BN) was constructed using imaging features and demographic data from literature.
• The BN's performance was evaluated on 99 thyroid nodules (54 benign, 45 malignant) with pathological confirmation.
• BN predictions were compared against assessments from two independent radiologists using ROC analysis.

Main Results:

• The BN demonstrated comparable performance to expert radiologists in distinguishing malignant from benign thyroid nodules.
• Area under the ROC curve (A(z)) values showed minimal differences between the BN and radiologists (e.g., 0.85 vs. 0.88).

Conclusions:

• A novel BN effectively integrates sonographic and demographic features to predict thyroid nodule malignancy.
• The developed BN offers a reliable tool for assessing malignancy risk, performing on par with expert radiologists.