Related Experiment Video
Updated: Sep 12, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Nomogram integrating BRAF V600E, serum biomarkers, and ultrasound features for malignancy risk stratification in
Yu Gao1, Wenran Zhang1, Jingmin Li1
1Department of Endocrinology, The Fourth Affiliated Hospital of Anhui Medical University, Hefei, China.
Background:
For patients with C-TIRADS 3 or 4 nodules who have undergone fine-needle aspiration cytology (FNAC), subsequent management decisions remain challenging, especially when cytological results are discordant with imaging findings or indeterminate. We built a nomogram incorporating BRAF V600E, serum biomarkers, and ultrasound features to provide an individualized risk estimate of malignancy for these nodules.
Methods:
We retrospectively included 300 patients with C-TIRADS 3 or 4 nodules. All nodules were confirmed by postoperative pathology or cytology. Among these, 76 were malignant and 224 benign. After running LASSO regression with 10-fold cross-validation and applying the 1-standard error rule to select core predictors, we built the nomogram using Firth-penalized logistic regression. Internal validity was checked using bootstrap resampling with 1,000 iterations while the model's performance was evaluated using ROC curves, calibration (Hosmer-Lemeshow test, Brier score) and decision curve analysis (DCA). The additional predictive value beyond the C-TIRADS system was measured by the Net Reclassification Index (NRI) and the Integrated Discrimination Improvement (IDI).
Results:
Five independent predictors were finally selected: BRAF V600E mutation (OR = 36.455, 95% CI: 13.381-118.094), TgAb positivity (OR = 4.808, 95% CI: 2.289-10.486), microcalcifications (OR = 4.168, 95% CI: 1.880-9.424), aspect ratio > 1 (OR = 3.027, 95% CI: 1.285-7.115), and serum VEGF (per 100 pg/mL increase, OR = 3.798, 95% CI: 1.763-8.677). Each predictor was independently associated with malignancy (all P < 0.05). The area under the curve (AUC) of our nomogram reached 0.890 (95% CI: 0.845-0.935; bootstrap-corrected: 0.886), which was significantly higher than that of C-TIRADS alone (AUC = 0.746, P < 0.001). Reclassification analysis yielded a categorical NRI of 0.436 (95% CI: 0.316-0.551, P < 0.001) and an IDI of 0.263 (95% CI: 0.182-0.341, P < 0.001). Calibration was satisfactory: the Hosmer-Lemeshow χ 2 gave 4.61 (P = 0.799), and the Brier score was 0.104. DCA demonstrated a higher net benefit across threshold probabilities ranging from 2 to 90%.
Conclusion:
The nomogram demonstrated good discriminative performance in C-TIRADS 3 or 4 nodules, supporting its utility as a post-FNAC decision aid.