Objective:

To investigate the application of ultrasound image and mathematical models in the diagnosis of benign and malignant partially cystic thyroid nodules.

Methods:

A retrospective study was conducted in 97 patients with thyroid nodules from Jan. 2018 to J

ul. 2019 in the First Affiliated Hospital of Nanjing Medical University. Each gray scale value of the 2-D ultrasound image and color Doppler flow imaging of thyroid nodule was read. There existed totally 97 nodes

including 40 benign and 57 malignant partially cystic thyroid nodules

respectively. The malignant partially-cystic thyroid nodules were confirmed by surgery and pathology

and benign partially-cystic thyroid nodules were confirmed by fine needle aspiration biopsy (FNAB). The ultrasonic features of nodules were analyzed. Then based on the variables of ultrasonic features with significant correlation

the partial least squares-discriminant analysis (PLS-DA) and logistic regression were used to diagnose partially cystic thyroid nodules.

Results:

Based on Spearman correlation analysis and Stepwise regression analysis

eight variables from ultrasonic features were selected

including age

position of cystic portion

calcification

margin

shape

orientation

and distribution of cystic components (limited or not)

and whether the thyroid is accompanied by diffuse lesions. According to the selected variables

the true positive and negative rates of PLS-DA were 92.9% and 96.6%

respectively. The true positive and negative rates of logistic regression were 92.0% and 88.8%

respectively. The area under curve (AUC) of PLS-DA was 0.947

which was much higher than that of logistic regression (0.904).

Conclusion:

The methods of PLS-DA and logistic regression coupled with ultrasound image could be used for diagnosis ofbenign and malignant partially cystic thyroid nodules. PLS-DA seems to be more powerful than ogistic regression.