PRS enhances ultrasound-based assessment of thyroid nodules

A recent study suggests that an ultrasound-based classification of thyroid nodules as either benign or malignant can improve its diagnostic accuracy by combining it with polygenic risk scores (PRS).

“Thyroid nodule ultrasound-based risk stratification schemas rely on the presence of high-risk sonographic features,” the authors said. “However, some malignant thyroid nodules have benign appearance on thyroid ultrasound.”

To address this, the authors evaluated PRS accounting for inherited thyroid cancer risk and combined this with ultrasound-based analysis. They trained the convolutional neural network classifier on thyroid ultrasound still images and cine clips from 621 thyroid nodules.

Additionally, phenome-wide association study (PheWAS) and PRS PheWAS were used to adjust PRS for the classification of malignant and benign nodules. A total of 73,346 participants in the Colorado Center for Personalized Medicine Biobank underwent PRS assessment.

Combining the deep learning model output with thyroid cancer PRS and genetic ancestry estimates augmented the area under the receiver operating characteristic curve (AUROC) of the benign versus malignant thyroid nodule classifier from 0.83 to 0.89 (DeLong: p=0.007).

The combination of deep learning and genetic classifier had a sensitivity of 0.95 (95 percent confidence interval [CI], 0.88‒0.99), a specificity of 0.63 (95 percent CI, 0.55‒0.70), a positive predictive value of 0.47 (95 percent CI, 0.41‒0.58), and a negative predictive value of 0.97 (95 percent CI, 0.92‒0.99).

The improvement seen in AUROC supported that of the European ancestry-stratified analysis: 0.83 for deep learning and 0.87 for combined deep learning and PRS classifiers).

Notably, elevated PRS significantly correlated with a higher risk of thyroid cancer structure disease recurrence (ordinal logistic regression: p=0.002).