Radiotherapy and oncology : journal of the European Society for Therapeutic Radiology and Oncology · 2025
Deep learning dosiomics in grade 4 radiation-induced lymphopenia prediction in radiotherapy for esophageal cancer: a multi-center study.
СтатьяМногоцентровое исследованиеПоддержка исследования: вне правительства СШАНазвание 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.
Zheng Qiao, Yu Xianwen, Pan Wenkai, Ai Yao, Liu Ziyang, Zhang Tongbin, Xiang Ziqing, Lei Xiyao, Zhu Mingqi, Wu Jianping, Xiao Chengjian, Shao Li, Guo Wenhao, Li Sunjun, Ma Junhao, Lou Yuxin, Liu Lijing, Xie Congying, Jin Xiance
Аннотация
PURPOSE: To investigate the feasibility and accuracy of using deep learning and dosiomics features, as well as their combination with dose-volume histogram (DVH) parameters and clinical factors to predict grade 4 radiation-induced lymphopenia (G4RIL) for patients with esophageal cancer (EC) who undergoing radiotherapy (RT).
METHODS: This retrospective study enrolled 545 patients with EC who underwent RT between January 2015 and December 2023 from five medical centers, and divided them into a training set, an internal validation set, an external test set 1, and an external test set 2, respectively. Dosiomics (D) and deep learning dosiomics (DLD) models were built to predict the probability of G4RIL based on radiation dose distributions using five-fold cross-validation. DVH parameters were extracted from organs-at-risk to build a G4RIL prediction model after dimensionality reduction using principal component analysis. A combination model integrating dosiomics, DLD features, DVH parameters and clinical factors (C) was investigated.
RESULTS: The D + DLD model and D + DLD + DVH model achieved mean area under curves (AUCs) of 0.78 ± 0.02 vs. 0.83 ± 0.02, 0.75 ± 0.04 vs. 0.80 ± 0.02, 0.77 ± 0.04 vs. 0.79 ± 0.03, and 0.70 ± 0.02 vs. 0.76 ± 0.03 in the training set, internal validation set, external test set 1, external test set 2, respectively. The combination model of D + DLD + DVH + C achieved the best predictive performance in the prediction of G4RIL with a mean AUC of 0.86 ± 0.03, 0.83 ± 0.03, 0.82 ± 0.04 and 0.78 ± 0.04 in the training set, internal validation set, external test set 1 and 2, respectively.
CONCLUSIONS: The combination model demonstrated the ability to effectively predict G4RIL in patients with EC undergoing RT.
METHODS: This retrospective study enrolled 545 patients with EC who underwent RT between January 2015 and December 2023 from five medical centers, and divided them into a training set, an internal validation set, an external test set 1, and an external test set 2, respectively. Dosiomics (D) and deep learning dosiomics (DLD) models were built to predict the probability of G4RIL based on radiation dose distributions using five-fold cross-validation. DVH parameters were extracted from organs-at-risk to build a G4RIL prediction model after dimensionality reduction using principal component analysis. A combination model integrating dosiomics, DLD features, DVH parameters and clinical factors (C) was investigated.
RESULTS: The D + DLD model and D + DLD + DVH model achieved mean area under curves (AUCs) of 0.78 ± 0.02 vs. 0.83 ± 0.02, 0.75 ± 0.04 vs. 0.80 ± 0.02, 0.77 ± 0.04 vs. 0.79 ± 0.03, and 0.70 ± 0.02 vs. 0.76 ± 0.03 in the training set, internal validation set, external test set 1, external test set 2, respectively. The combination model of D + DLD + DVH + C achieved the best predictive performance in the prediction of G4RIL with a mean AUC of 0.86 ± 0.03, 0.83 ± 0.03, 0.82 ± 0.04 and 0.78 ± 0.04 in the training set, internal validation set, external test set 1 and 2, respectively.
CONCLUSIONS: The combination model demonstrated the ability to effectively predict G4RIL in patients with EC undergoing RT.