MRI
‐Derived Lymph Node Characteristics for Risk Stratification of Distant Metastasis in Locoregionally Advanced Nasopharyngeal Carcinoma
Jing Wang, Yuxing Yu, Honghong Zhang, Haosheng Zhang, Jianxian Chen, Jiajia Zheng, Qisi Zhang, Hengyu Guo, Lanxin Lin, Yubin Hu, Sufang Qiu ABSTRACT
Background
MRI‐derived nodal features may improve prognostication beyond TNM staging, but distant‐metastasis risk models remain limited.
Purpose
To develop and internally validate an MRI model for distant metastasis prediction in locoregionally advanced nasopharyngeal carcinoma (LA‐NPC).
Study Type
Retrospective.
Population
One thousand five‐hundred and forty‐one patients with LA‐NPC (1103 men and 438 women; mean age, 48 ± 12 years) were split into a training set ( n = 1079) and an internal validation set ( n = 462).
Field Strength/Sequence
3.0‐T MRI using axial T1‐weighted turbo spin‐echo, axial T2‐weighted short‐tau inversion recovery, diffusion‐weighted imaging ( b = 0 and 800 s/mm 2 ) with apparent diffusion coefficient maps, and multiplanar fat‐suppressed contrast‐enhanced T1‐weighted turbo spin‐echo.
Assessment
Extranodal extension (ENE), cervical lymph node necrosis (CNN), nodal size, regional lymph node density (RLND), nodal laterality, and clinical variables were assessed. Recursive partitioning analysis (RPA) defined risk groups, which were combined with T and N categories in a nomogram.
Statistical Tests
Cox analyses were performed in training. Performance was assessed by the concordance index (C‐index), 3‐year time‐dependent receiver operating characteristic analysis, and calibration. Two‐sided p < 0.05 indicated significance.
Results
ENE Grade 3 (HR, 1.92; 95% CI, 1.20–3.07), CNN Grades 3–4 (HR, 2.65; 95% CI, 1.46–4.81), and RLND (per‐unit increase: HR, 1.92; 95% CI, 1.29–2.88) were independently associated with shorter distant metastasis–free survival (DMFS). The multivariable model C‐index was 0.766 (95% CI, 0.721–0.811). RPA assigned 491, 524, and 64 patients to low‐, intermediate‐, and high‐risk groups, with 3‐year DMFS rates of 97.53%, 89.90%, and 60.83%, respectively. Nomogram 3‐year AUCs were 0.777 (95% CI, 0.725–0.828) in training and 0.728 (95% CI, 0.645–0.810) in validation; the difference was nonsignificant ( p = 0.333). Mean absolute calibration errors were 0.009 and 0.012, respectively.
Data Conclusion
MRI‐derived nodal features enabled DMFS risk stratification, and the internally validated nomogram showed discrimination and calibration.
Evidence Level
4.
Technical Efficacy
Stage 2.