Model-free parameters derived from intravoxel inherent motion combined with tumor aggressiveness indicators in
Fan Yang1, Wenjing Peng2, Haoran Wei2
1Department of Radiology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing 100021, China; Department of Radiology, Division of Neuroimaging and Neurointervention, Stanford University, Stanford University Medical Center, Stanford, California, USA.
Purpose:
To evaluate the value of model-free parameters (virtual magnetic resonance elastography [vMRE] and signature index [S-index]) and tumor aggressiveness indicators (contralateral pharyngeal recess invasion [CPRI] and tumor growth type [TGT]) in forecasting 5-year treatment outcomes in nasopharyngeal carcinoma (NPC).
Materials And Methods:
A total of 106 patients were included in this study. Only two b-values were employed to calculate vMRE and S-index. Univariate and multivariate Cox analyses were conducted with clinical factors as confounding variables. Multivariate logistic analysis was used to construct multiple models. The additional value of CPRI and TGT was evaluated through net reclassification improvement index (NRI) and integrated discrimination improvement (IDI). Model performance and robustness were evaluated.
Results:
The predictive performance of vMRE and S-index (area under the curve [AUC]: 0.620 ∼ 0.639) was comparable to that of model-based parameters (AUC: 0.570 ∼ 0.658, all P ≥ 0.366). CPRI and TGT independently predicted progression-free survival (PFS) and overall survival (OS). Combined models significantly outperformed than the TNM stage model (all P ≤ 0.002) and vMRE-based or S-index-based model (all NRI ≥ 0.105 and IDI ≥ 0.085) in terms of 5-year OS (AUC: 0.840 ∼ 0.843) and 5-year PFS (AUC: 0.786 ∼ 0.793). The combined model demonstrated robust performance through Hosmer-Lemeshow test, cross-validation (iteration = 1000), 7:3 validation (chronologically or randomly), and post-hoc subgroup analysis.
Conclusion:
The advent of vMRE and S-index has streamlined the clinical applications without compromising the value of IVIM technique. Incorporating CPRI and TGT has further enhanced the predictive performance.
Clinical Relevance Statement:
Only two b-values are needed to calculate vMRE and S-index, which, along with CPRI and TGT, are pivotal in forecasting long-term survival in patients with NPC.
