Computer methods and programs in biomedicine · 2026
Dual-branch attention-enhanced network integrating CT images and clinicoradiographic features for preoperative ternary classification of IASLC grading in lung adenocarcinoma: A multicenter study.
СтатьяМногоцентровое исследованиеНазвание 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.
Zuo Zhichao, Deng Jinqiu, Zeng Ying, Qi Wanyin, Liu Wen, Zhang Jianping
Аннотация
BACKGROUND AND OBJECTIVE: Precise preoperative prediction of the International Association for the Study of Lung Cancer (IASLC) grade in lung adenocarcinoma (LUAD) is essential for risk stratification and surgical decision-making. However, existing approaches remain limited in reliably distinguishing among the three IASLC grades. We developed and validated a dual-branch attention-enhanced network that integrates computed tomography (CT) imaging and clinicoradiographic (CR) information for preoperative ternary IASLC grading.
METHODS: This retrospective multicenter study included 1477 thin-slice chest CT examinations from three independent institutions. The proposed framework, DB-CRN, jointly models a preprocessed 3D nodule subvolume and patient-specific CR features through a two-stage dual-branch architecture equipped with a convolutional block attention module and channel attention blocks. Three comparative models were constructed: DB-NoCRN (image-only ablation), ML-CR (CR-only XGBoost model), and a conventional radiomics pipeline. Model performance was evaluated using the macro-averaged area under the receiver operating characteristic curve (macro AUC) and the Obuchowski index.
RESULTS: DB-CRN achieved the best overall performance, with a macro AUC of 0.858 and an Obuchowski index of 0.844, outperforming DB-NoCRN (macro AUC 0.837; Obuchowski index 0.826), ML-CR (macro AUC 0.787; Obuchowski index 0.729), and the radiomics model (macro AUC 0.811; Obuchowski index 0.810).
CONCLUSION: By fusing multiscale CT representations with CR priors under an attention-enhanced dual-branch design, DB-CRN enables robust preoperative ternary classification of IASLC grades in LUAD. This framework offers a noninvasive and reproducible tool to support risk-adapted surgical planning and personalized management, warranting prospective validation.
METHODS: This retrospective multicenter study included 1477 thin-slice chest CT examinations from three independent institutions. The proposed framework, DB-CRN, jointly models a preprocessed 3D nodule subvolume and patient-specific CR features through a two-stage dual-branch architecture equipped with a convolutional block attention module and channel attention blocks. Three comparative models were constructed: DB-NoCRN (image-only ablation), ML-CR (CR-only XGBoost model), and a conventional radiomics pipeline. Model performance was evaluated using the macro-averaged area under the receiver operating characteristic curve (macro AUC) and the Obuchowski index.
RESULTS: DB-CRN achieved the best overall performance, with a macro AUC of 0.858 and an Obuchowski index of 0.844, outperforming DB-NoCRN (macro AUC 0.837; Obuchowski index 0.826), ML-CR (macro AUC 0.787; Obuchowski index 0.729), and the radiomics model (macro AUC 0.811; Obuchowski index 0.810).
CONCLUSION: By fusing multiscale CT representations with CR priors under an attention-enhanced dual-branch design, DB-CRN enables robust preoperative ternary classification of IASLC grades in LUAD. This framework offers a noninvasive and reproducible tool to support risk-adapted surgical planning and personalized management, warranting prospective validation.