Radiopathomics Model Predicts Progression-Free Survival in Nasopharyngeal Carcinoma
核心洞察
An integrative radiopathomics model, combining radiomics, pathomics, and clinical factors, shows promise in predicting progression-free survival (PFS) for nasopharyngeal carcinoma (搜索) (NPC) patients.
The model utilizes MRI (搜索) and whole-slide imaging (WSI) to extract predictive features, enhancing risk stratification and treatment planning for NPC.
The radiopathomics model demonstrated superior performance compared to radiomics, pathomics, or clinical models alone, as validated through concordance index and calibration curve analyses.
An integrative radiopathomics model has been developed to predict progression-free survival (PFS) in patients with nasopharyngeal carcinoma (搜索) (NPC). This novel approach combines radiomic features extracted from MRI (搜索) scans, pathomic features from whole-slide imaging (WSI) of tumor tissue, and independent clinical factors to provide a comprehensive prognostic assessment. The study highlights the potential of integrating multi-modal data to improve risk stratification and personalize treatment strategies for NPC patients.
The study retrospectively analyzed data from NPC patients, utilizing T2-weighted imaging (T2WI) and contrast-enhanced T1-weighted imaging (CET1-w) MRI (搜索) sequences, along with WSI of H&E-stained tumor samples. Radiomic features were extracted from MRI using AK software (搜索), while pathomic features were derived from WSI using a Swin Transformer deep learning architecture. These features were then combined with clinical parameters in a radiopathomics nomogram to predict PFS.
The radiomics signature was constructed using minimum redundancy maximum relevance (mRMR) and least absolute shrinkage and selection operator (LASSO) methods for feature selection, resulting in a Rad-score. The pathomics signature was generated using a Swin Transformer, a deep learning architecture, to extract features from WSI. Independent clinical prognostic factors were identified through univariate and multivariate Cox regression analyses.
The performance of the radiopathomics model was evaluated using the concordance index (C-index) and calibration curves. The radiopathomics model demonstrated a superior C-index compared to models based on radiomics, pathomics, or clinical factors alone, indicating improved prognostic accuracy. The calibration curve also showed good agreement between predicted and observed PFS rates. Risk stratification analysis further revealed that the radiopathomics model effectively discriminated between patient subgroups with different PFS outcomes.
The MRI (搜索) procedure utilized a 1.5-Tesla MRI scanner with specific protocols for T1WI, T2WI, and CET1-w images. Gadodiamide was administered intravenously for CET1-w imaging. Preprocessing of MRI data included resampling, skull stripping, and intensity standardization. Two radiologists independently performed segmentation of the primary NPC lesion, ensuring reproducibility through inter- and intraclass correlation coefficients (ICCs) > 0.75.
All patients underwent regular follow-ups with MRI (搜索) examinations at specified intervals. PFS was defined as the time from therapy initiation to disease progression, death, or the last follow-up visit. Treatment regimens included radiotherapy alone for stage I-II disease and concurrent chemoradiotherapy (CCRT) with or without adjuvant/induction chemotherapy for stage II-IVA disease. Radiotherapy was administered at a total dose of 70-76 Gy.
This integrative radiopathomics model offers a non-invasive approach to enhance prognostic assessment and personalize treatment strategies in NPC. By combining imaging and pathological data, the model provides a more comprehensive understanding of tumor biology and clinical behavior, potentially leading to improved patient outcomes.