Frontiers in oncology · 2026
Foundation model-enhanced multimodal radiomics for predicting response to chemo-immunotherapy in advanced lung squamous cell carcinoma.
СтатьяНазвание 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.
Wang Zhichao, Zhang Yang, He Chuchu, Wang Meng, Cai Jun
Аннотация
OBJECTIVE: To develop and validate a foundation model-driven multimodal fusion framework for non-invasive prediction of treatment response to first-line chemo-immunotherapy in patients with advanced lung squamous cell carcinoma (LUSC).
METHODS: In this retrospective study, baseline contrast-enhanced computer tomography (CT) images and clinical data from patients with advanced LUSC receiving first-line chemo-immunotherapy were collected. Handcrafted radiomics features were extracted from tumor regions of interest, and 2.5D deep learning features were extracted using a DINO-pretrained vision Transformer foundation model. Clinical variables were incorporated to construct a multi-source features fusion model based on machine learning classifiers. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. DeLong testing, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI) were performed for comparative assessment. Shapley additive explanations (SHAP) analysis was applied to enhance model interpretability and decision transparency.
RESULTS: Among the constructed models, the multi-source features fusion model (FusionModel) achieved the best predictive performance, with an AUC of 0.903 and an accuracy of 0.885 in the training cohort, and an AUC of 0.863 and an accuracy of 0.836 in the validation cohort. FusionModel outperformed models based on single-modality features and demonstrated superior net clinical benefit on DCA. NRI and IDI analyses further supported improved reclassification ability. SHAP analysis revealed that deep learning features contributed dominantly, while radiomics and clinical variables provided complementary prognostic information, supporting the biological plausibility of the model.
CONCLUSION: The foundation model-driven multi-source features fusion model enabled accurate and interpretable prediction of chemo-immunotherapy response in advanced LUSC. This strategy demonstrated strong discriminative performance and clinical applicability, highlighting its potential as a non-invasive tool for individualized treatment stratification.
METHODS: In this retrospective study, baseline contrast-enhanced computer tomography (CT) images and clinical data from patients with advanced LUSC receiving first-line chemo-immunotherapy were collected. Handcrafted radiomics features were extracted from tumor regions of interest, and 2.5D deep learning features were extracted using a DINO-pretrained vision Transformer foundation model. Clinical variables were incorporated to construct a multi-source features fusion model based on machine learning classifiers. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity. DeLong testing, decision curve analysis (DCA), net reclassification improvement (NRI), and integrated discrimination improvement (IDI) were performed for comparative assessment. Shapley additive explanations (SHAP) analysis was applied to enhance model interpretability and decision transparency.
RESULTS: Among the constructed models, the multi-source features fusion model (FusionModel) achieved the best predictive performance, with an AUC of 0.903 and an accuracy of 0.885 in the training cohort, and an AUC of 0.863 and an accuracy of 0.836 in the validation cohort. FusionModel outperformed models based on single-modality features and demonstrated superior net clinical benefit on DCA. NRI and IDI analyses further supported improved reclassification ability. SHAP analysis revealed that deep learning features contributed dominantly, while radiomics and clinical variables provided complementary prognostic information, supporting the biological plausibility of the model.
CONCLUSION: The foundation model-driven multi-source features fusion model enabled accurate and interpretable prediction of chemo-immunotherapy response in advanced LUSC. This strategy demonstrated strong discriminative performance and clinical applicability, highlighting its potential as a non-invasive tool for individualized treatment stratification.