Predicting high-burden central compartment lymph node metastasis (≥ 3 positive nodes) in papillary thyroid carcinoma: a habitat radiomic framework based on reconstructed contrast-enhanced CT
- 1Affiliated Hospital of Qingdao University
Abstract
Background Lymph node metastasis is common in papillary thyroid carcinoma and increases the risk of locoregional recurrence. Preoperative identification of a clinically actionable nodal burden could improve treatment planning, but conventional imaging and clinical variables discriminate poorly. We developed a habitat radiomic framework to separate high-burden central compartment nodal disease (≥ 3 pathologically positive nodes) from low-burden disease (0–2 positive nodes). Methods We retrospectively included 119 patients with pathologically confirmed papillary thyroid carcinoma from two independent institutions: a training cohort of 71 patients and an independent-centre test cohort of 48 patients. Contrast-enhanced CT images underwent in-plane super-resolution reconstruction. The thyroid gland was partitioned into imaging-defined habitat subregions, from which 428 radiomic features were extracted. All feature selection and model development were performed in the training cohort only; the unsupervised habitat clustering was fitted on the pooled voxels of the whole cohort, which is reported as a limitation. An XGBoost model was compared with a logistic regression clinical model built on nine preoperative variables. Results Four habitat subregions were identified. The habitat-only model achieved a test AUC of 0.913 (95% CI 0.831–0.994), numerically higher than the nine-variable clinical model (0.731, 95% CI 0.585–0.876; difference in AUC +0.182, p = 0.040). Against a parsimonious age-and-sex model (AUC 0.777) the difference was not statistically significant (p = 0.105). Training performance reached an AUC of 1.000, indicating substantial overfitting. Conclusions Habitat radiomics of reconstructed contrast-enhanced CT showed numerically higher discrimination than a nine-variable clinical comparator in this test cohort, although the advantage over age and sex alone was not statistically significant and no comparison was adjusted for multiplicity. These exploratory findings require prospective multicentre validation.
Related
- The clinical value of ultrasound radiomics combined with clinical data nomogram in predicting cervical lymph node metastasis of papillary thyroid carcinoma.
- Multimodal radiomics based on dynamic contrast-enhanced ultrasound for predicting cervical lymph node metastasis in papillary thyroid carcinoma with Hashimoto’s thyroiditis
- Treatment strategies for unilateral papillary thyroid carcinoma with ipsilateral cervical lymph node metastasis: a narrative review
- Machine learning-based prediction of lymph node metastasis in papillary thyroid carcinoma: an interpretable multidimensional model
- Combined apparent diffusion coefficient histogram and MRI-based fractal analysis for predicting lymph node metastasis in esophageal squamous cell carcinoma
BibTeX
@article{sun2026predicting,
title = {{Predicting high-burden central compartment lymph node metastasis (≥ 3 positive nodes) in papillary thyroid carcinoma: a habitat radiomic framework based on reconstructed contrast-enhanced CT}},
author = {Dawei Sun and Huichao Li and Yanmei Zhan and Zhoujing Zhang},
journal = {The Egyptian Journal of Radiology and Nuclear Medicine},
year = {2026},
month = oct,
doi = {10.1186/s43055-026-01894-x},
url = {https://doi.org/10.1186/s43055-026-01894-x},
note = {CC BY},
}
RIS
TY - JOUR
TI - Predicting high-burden central compartment lymph node metastasis (≥ 3 positive nodes) in papillary thyroid carcinoma: a habitat radiomic framework based on reconstructed contrast-enhanced CT
AU - Dawei Sun
AU - Huichao Li
AU - Yanmei Zhan
AU - Zhoujing Zhang
PY - 2026
DA - 2026/10/25
JO - The Egyptian Journal of Radiology and Nuclear Medicine
DO - 10.1186/s43055-026-01894-x
UR - https://doi.org/10.1186/s43055-026-01894-x
AB - Background Lymph node metastasis is common in papillary thyroid carcinoma and increases the risk of locoregional recurrence. Preoperative identification of a clinically actionable nodal burden could improve treatment planning, but conventional imaging and clinical variables discriminate poorly. We developed a habitat radiomic framework to separate high-burden central compartment nodal disease (≥ 3 pathologically positive nodes) from low-burden disease (0–2 positive nodes). Methods We retrospectively included 119 patients with pathologically confirmed papillary thyroid carcinoma from two independent institutions: a training cohort of 71 patients and an independent-centre test cohort of 48 patients. Contrast-enhanced CT images underwent in-plane super-resolution reconstruction. The thyroid gland was partitioned into imaging-defined habitat subregions, from which 428 radiomic features were extracted. All feature selection and model development were performed in the training cohort only; the unsupervised habitat clustering was fitted on the pooled voxels of the whole cohort, which is reported as a limitation. An XGBoost model was compared with a logistic regression clinical model built on nine preoperative variables. Results Four habitat subregions were identified. The habitat-only model achieved a test AUC of 0.913 (95% CI 0.831–0.994), numerically higher than the nine-variable clinical model (0.731, 95% CI 0.585–0.876; difference in AUC +0.182, p = 0.040). Against a parsimonious age-and-sex model (AUC 0.777) the difference was not statistically significant (p = 0.105). Training performance reached an AUC of 1.000, indicating substantial overfitting. Conclusions Habitat radiomics of reconstructed contrast-enhanced CT showed numerically higher discrimination than a nine-variable clinical comparator in this test cohort, although the advantage over age and sex alone was not statistically significant and no comparison was adjusted for multiplicity. These exploratory findings require prospective multicentre validation.
SN - 2090-4762
SN - 0378-603X
ER -