Related Experiment Video
Updated: May 5, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Prospective Validation of an Ultrasound Capsule-Based Model for Predicting Follicular Thyroid Carcinoma Without
Xiang-Yun Yao1, Xin Li2, Fang Mei3
1Department of Ultrasound, Peking University Third Hospital, Beijing, China.
Background:
The surgical indications for follicular thyroid neoplasms (FTNs) remain controversial due to challenges in the preoperative follicular thyroid carcinoma (FTC) diagnosis. We aimed to explore the sonographic features of the FTN capsule and establish a prediction model for diagnosing FTC without high-risk features.
Methods:
This prospective cohort study enrolled consecutive adult patients with FTN. Patients presenting with extrathyroidal extension or extracapsular angioinvasion on preoperative imaging were excluded. Intraoperative ultrasound (US)-guided incisions were conducted in 20 patients during thyroidectomy. Sonographic features of FTN capsules were identified and validated through comparison with US, macroscopic, and microscopic pathology images from the same US-selected section. Invaded capsules were categorized based on pathological indicators of capsular invasion. The diagnostic performance of the US capsule-based model and US risk stratification systems (American College of Radiology Thyroid Imaging Reporting and Data System [ACR-TIRADS] and Ultrasound Follicular Thyroid Imaging Reporting and Data System [F-TIRADS]) were compared.
Results:
Seventy-four patients with unifocal lesions and 14 patients with multifocal lesions were enrolled and pathologically diagnosed with follicular thyroid adenoma (n = 67) and FTC (n = 35). As widely invasive FTC was initially excluded, there were 34 minimally invasive subtypes and 1 encapsulated angioinvasive subtype. The areas under the curves for the US capsule-based model, ACR-TIRADS, and F-TIRADS were 0.839 (95% confidence interval [CI], 0.753-0.904), 0.852 (95% CI, 0.768-0.914), and 0.840 (95% CI, 0.755-0.905), respectively. No significant differences were observed in the areas under the curves, sensitivity, or accuracy of the US capsule-based model, ACR-TIRADS, and F-TIRADS. The specificities of the US capsule-based model and F-TIRADS were higher than that of the ACR-TIRADS (88.1% and 80.60%, respectively, vs. 44.8%, p < 0.05).
Conclusions:
Careful US scanning enables clear FTN capsule visualization, providing a straightforward, specific method for diagnosing FTC and guiding completion thyroidectomy by detecting intracapsular angioinvasion in patients with FTC.
