Ultrasound radiomics for predicting extrathyroidal extension in children

Extrathyroidal extension (ETE) has strong predictive ability based on ultrasound radiomics for pediatric papillary thyroid cancer (PTC), according to a recent study.

Pediatric PTC has been associated with greater aggressiveness, along with increased susceptibility to ETE, lymph node metastasis (LNM), and distant metastasis. A primary tumor with a size over 4 cm in the thyroid gland is minimal ETE, while extensive ETE is defined as, “the primary tumor invasion of subcutaneous soft tissue, trachea, larynx, esophagus, recurrent laryngeal nerve, carotid artery, prevertebral fascia, or mediastinal vessels.”

ETE is defined as an independent risk factor for LNM and is associated with distant metastases and higher tumor recurrence. The most common method for preoperatively examining thyroid cancer is ultrasound, but data on ultrasonomics of papillary thyroid carcinoma in children are limited.

To determine the effectiveness of ultrasound radiomics for predicting ETE in papillary thyroid cancer in children, investigators conducted a study from January 2013 through August 2022. Participants were recruited from the Cancer Institute and Hospital at Tianjin Medical University.

Inclusion criteria included being diagnosed with postoperative pathology if they had PTC, having ultrasound examination and surgery at Tianjin Medical University Cancer Hospital, and being aged under 18 years. Data was gathered on participants’ sex and age alongside clinical data.

Ultrasound evaluation occurred before surgery, with patients lying supine with their head moderately tilted. The size of thyroid tumors, tumor location, internal echo pattern, tumor boundary, and tumor calcification were observed.

Two radiologists independently analyzed ultrasound imaging without knowledge of historical findings. ETE diagnosis criteria included more than 25% of the lesion’s circumference being in contact with the thyroid capsule or envelope line, or a tumor of any size exceeding the thyroid capsule.

Radiomics features were determined through areas of interest along the perimeter of the tumor contour. The correlation coefficient screening method was then used to eliminate coefficients, reducing the feature dimension.

Investigators developed 4 prediction models based on classifiers, with model performance compared through the receiver operating characteristic (ROC) curves. The area under the curve (AUC) was measured for ROC curves to allow comparison. K-nearest neighbor (KNN), random forest, support vector machine (SVM), and LightGBM were the models developed.

The mean age of patients was 14.6 years, and the number of female patients more than doubled male patients. Of the children, 103 were identified as ETE and 61 non-ETE. A training group consisted of 115 patients and a validation group of 49.

In the training cohort, the mean AUC was 0.88 for SVM, 0.873 for KNN, 0.999 for random forest, and 0.926 for LightGBM. In the validation cohort, this was 0.784 for SVM, 0.72 for KNN, 0.728 for random forest, and 0.832 for LightGBM.

LightGBM showed efficiency overall in both groups. Overall, ultrasonic radiomics showed strong predictive ability for ETE in pediatric PTC.

Reference

Li J, Xia F, Wang X, et al. Multiclassifier radiomics analysis of ultrasound for prediction of extrathyroidal extension in papillary thyroid carcinoma in children. Int J Med Sci. 2023;20(2):278-286. doi:10.7150/ijms.79758