European journal of radiology · 2025
Interpretable combined models for predicting treatment response and hematologic toxicity in locally advanced gastric cancer treated with PD-1 blockade and neoadjuvant chemotherapy.
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.
Wei Yuze, Zhu Yanmei, Dong Qian, Wang Wentao, Yu Tao, Zhang Jianjun, Dong Yue
Аннотация
OBJECTIVES: The aim of this study was to establish and validate the interpretable combined models for treatment response (TR) and hematologic toxicity (HT) after programmed cell death protein 1 blockade plus neoadjuvant chemotherapy in locally advanced gastric cancer (LAGC).
METHODS: 195 patients with LAGC were enrolled from two centres (117, 50 and 28 patients in the training, internal validation and external validation cohorts). The changes of radiomics features from pre-treatment and post-treatment gastric computed tomography images were extracted to build delta radiomics score and predict TR, as well as another radiomics score from pre-treatment vertebral computed tomography images for the prediction of HT, and both combined models were established from clinicopathological characteristics and radiomics score. The predictive performance of the combined models were evaluated using receiver operating characteristics, calibration and decision curve analyses. Interpretability was assessed using the shapley additive explanations framework (SHAP).
RESULTS: The area under curve (AUC) in the combined model for predicting TR were 0.893,0.846 and 0.913 for the training, internal validation and external validation cohorts, respectively. The AUC in the combined model for predicting HT were 0.871, 0.917 and 0.865 for the training, internal validation and external validation cohorts, respectively. SHAP summary plots show the importance of each feature on the prediction outcome, while waterfall and force plots depict individual features' contributions to a response variable.
CONCLUSIONS: The combined models based on radiomics features for predicting TR and HT after programmed cell death protein 1 blockade plus neoadjuvant chemotherapy in LAGC demonstrated good predictive performance.
METHODS: 195 patients with LAGC were enrolled from two centres (117, 50 and 28 patients in the training, internal validation and external validation cohorts). The changes of radiomics features from pre-treatment and post-treatment gastric computed tomography images were extracted to build delta radiomics score and predict TR, as well as another radiomics score from pre-treatment vertebral computed tomography images for the prediction of HT, and both combined models were established from clinicopathological characteristics and radiomics score. The predictive performance of the combined models were evaluated using receiver operating characteristics, calibration and decision curve analyses. Interpretability was assessed using the shapley additive explanations framework (SHAP).
RESULTS: The area under curve (AUC) in the combined model for predicting TR were 0.893,0.846 and 0.913 for the training, internal validation and external validation cohorts, respectively. The AUC in the combined model for predicting HT were 0.871, 0.917 and 0.865 for the training, internal validation and external validation cohorts, respectively. SHAP summary plots show the importance of each feature on the prediction outcome, while waterfall and force plots depict individual features' contributions to a response variable.
CONCLUSIONS: The combined models based on radiomics features for predicting TR and HT after programmed cell death protein 1 blockade plus neoadjuvant chemotherapy in LAGC demonstrated good predictive performance.