BMC medical imaging · 2026
An explainable multimodal 2.5D deep learning-radiomics model for predicting extranodal extension in lung adenocarcinoma using preoperative CT: a multicenter retrospective cohort study.
СтатьяНазвание и аннотация: Перевод в очереди
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.
Wang Jianyu, Wang Yuhang, Zhao Xiaojiang, Zhang Han, Mu Qiuqiao, Wang Yanbo, Li Xin, Xu Meilin, Song Zhenchun, Ren Jie, Wang Zixiao, Sun Daqiang
Аннотация
OBJECTIVE: This study aimed to develop the Radiomics-Assembled ENE system (RAIEm), a multimodal preoperative computed tomography (CT) radiomics model, for identifying extranodal extension (ENE) in lung adenocarcinoma (LUAD) and to examine its utility in risk stratification and individualized treatment planning.
METHODS: We retrospectively assembled preoperative thin-slice chest CT scans and postoperative hematoxylin-eosin-stained histology from patients with LUAD. Tumors were segmented automatically using a 3D U-Net with virtual adversarial training. Radiomic features were derived from the intratumoral region and from concentric peritumoral rings expanded by 2, 4, 6, and 8 mm, and 2.5D deep-learning features were extracted from five axial CT slices centered on the largest tumor section using a convolutional neural network. Subsequently, we constructed an integrated Radiomics-ENE model (RAIEm), whose performance was assessed in separate training and validation cohorts.
RESULTS: RAIEm outperformed conventional unimodal models, yielding higher area under the receiver operating characteristic curve (AUC), sensitivity, and specificity in both the training and external validation cohorts. In the validation cohort, AUC was 0.78 (95% CI, 0.66-0.89), and decision-curve analysis indicated greater net benefit across most threshold probabilities. RAIEm demonstrated significant discriminative power for recurrence-free survival risk stratification in both the training and external validation cohorts (both p < 0.01). ENE-positive patients showed significantly shorter recurrence-free survival than ENE-negative patients in both cohorts. Shapley Additive Explanations (SHAP) analysis highlighted key radiomic drivers, with deep-learning features (one 2D and two 2.5D) ranking among the top three contributors.
CONCLUSION: RAIEm - a multimodal, preoperative CT radiomics-based model - showed encouraging performance for identifying ENE in LUAD and may support preoperative risk assessment; however, prospective multicenter validation is required before routine clinical implementation.
METHODS: We retrospectively assembled preoperative thin-slice chest CT scans and postoperative hematoxylin-eosin-stained histology from patients with LUAD. Tumors were segmented automatically using a 3D U-Net with virtual adversarial training. Radiomic features were derived from the intratumoral region and from concentric peritumoral rings expanded by 2, 4, 6, and 8 mm, and 2.5D deep-learning features were extracted from five axial CT slices centered on the largest tumor section using a convolutional neural network. Subsequently, we constructed an integrated Radiomics-ENE model (RAIEm), whose performance was assessed in separate training and validation cohorts.
RESULTS: RAIEm outperformed conventional unimodal models, yielding higher area under the receiver operating characteristic curve (AUC), sensitivity, and specificity in both the training and external validation cohorts. In the validation cohort, AUC was 0.78 (95% CI, 0.66-0.89), and decision-curve analysis indicated greater net benefit across most threshold probabilities. RAIEm demonstrated significant discriminative power for recurrence-free survival risk stratification in both the training and external validation cohorts (both p < 0.01). ENE-positive patients showed significantly shorter recurrence-free survival than ENE-negative patients in both cohorts. Shapley Additive Explanations (SHAP) analysis highlighted key radiomic drivers, with deep-learning features (one 2D and two 2.5D) ranking among the top three contributors.
CONCLUSION: RAIEm - a multimodal, preoperative CT radiomics-based model - showed encouraging performance for identifying ENE in LUAD and may support preoperative risk assessment; however, prospective multicenter validation is required before routine clinical implementation.