Academic radiology · 2026
Cluster-Based MR Radiomics Model for Predicting Induction Chemotherapy Response in Nasopharyngeal Carcinoma.
СтатьяПоддержка исследования: вне правительства СШАНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Huang Zhenhuan, Tu Xuezhao, Cao Lihang, Qiu Jing, You Qikui, Lin Dandan, Yu Tingyu, Ma Hui, Li Yueming
Аннотация
RATIONALE AND OBJECTIVES: To develop a cluster-specific magnetic resonance (MR) radiomics model for predicting induction chemotherapy (ICT) response in nasopharyngeal carcinoma (NPC) with enhanced interpretability using Shapley Additive exPlanations (SHAP).
MATERIALS AND METHODS: In this single-center, retrospective study, 225 patients with NPC were randomly assigned to the training (n = 158) and test (n = 67) cohorts. Tumor burden reduction ratio (TBRR), derived from pre- and post-ICT MR images, was used to classify patients as responders or non-responders. Tumor volumes of interest were subdivided into three radiomics-defined clusters, and radiomics features were extracted from each cluster and the whole tumor. Dimensionality reduction was performed using the intra-class correlation coefficient, Pearson correlation coefficient, and recursive feature elimination. Cluster-specific, whole-tumor, and multicluster radiomics models were constructed using six machine learning classifiers. Model performance was assessed by area under the curve (AUC), calibration, and decision curve analysis. SHAP was applied to interpret feature contributions at the cohort and individual levels.
RESULTS: Cluster 1, representing the vascularized tumor periphery, was the largest and most stable subregion and showed the strongest correlation between volume reduction and overall TBRR (R = 0.726, P < 0.001). The cluster 1-specific support vector machine model achieved the best performance (AUCs of 0.812 and 0.800 in the training and test cohorts, respectively), with good calibration and clinical utility. SHAP highlighted features reflecting intratumoral heterogeneity and voxel intensity, explaining inter-patient variability.
CONCLUSION: The cluster 1-specific MR radiomics model reliably predicted the ICT response in NPC and may offer interpretable prognostic insights regarding treatment beneficiaries.
MATERIALS AND METHODS: In this single-center, retrospective study, 225 patients with NPC were randomly assigned to the training (n = 158) and test (n = 67) cohorts. Tumor burden reduction ratio (TBRR), derived from pre- and post-ICT MR images, was used to classify patients as responders or non-responders. Tumor volumes of interest were subdivided into three radiomics-defined clusters, and radiomics features were extracted from each cluster and the whole tumor. Dimensionality reduction was performed using the intra-class correlation coefficient, Pearson correlation coefficient, and recursive feature elimination. Cluster-specific, whole-tumor, and multicluster radiomics models were constructed using six machine learning classifiers. Model performance was assessed by area under the curve (AUC), calibration, and decision curve analysis. SHAP was applied to interpret feature contributions at the cohort and individual levels.
RESULTS: Cluster 1, representing the vascularized tumor periphery, was the largest and most stable subregion and showed the strongest correlation between volume reduction and overall TBRR (R = 0.726, P < 0.001). The cluster 1-specific support vector machine model achieved the best performance (AUCs of 0.812 and 0.800 in the training and test cohorts, respectively), with good calibration and clinical utility. SHAP highlighted features reflecting intratumoral heterogeneity and voxel intensity, explaining inter-patient variability.
CONCLUSION: The cluster 1-specific MR radiomics model reliably predicted the ICT response in NPC and may offer interpretable prognostic insights regarding treatment beneficiaries.