Physics and imaging in radiation oncology · 2025
Integrating dose-volume histogram parameters and radiomics-based machine learning to identify carbon ion radiotherapy-induced acute oral mucositis in patients with head and neck cancer.
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.
Meng Xiangdi, Ju Zhuojun, Sakai Makoto, Li Yang, Musha Atsushi, Kubo Nobuteru, Kawamura Hidemasa, Ohno Tatsuya
Аннотация
BACKGROUND AND PURPOSE: Acute oral mucositis (AOM) is a critical adverse event in head and neck cancer (HNC) requiring early intervention during carbon ion radiotherapy (CIRT). Considering external irradiation and internal mucosal heterogeneity, this study developed a classification model integrating maximum dose (Dmax) and dose-volume-based radiomics (RaVx) for early identification of grade ≥ 2 AOM.
METHODS AND MATERIALS: We retrospectively analyzed 190 HNC patients treated with CIRT (training: test = 7:3). Radiomic features were extracted from the entire oral mucosa and dose-specific subregions [RaVx, 10-50 Gy(RBE) in 10-Gy increments]. Feature selection employed logistic regression, random forest, and XGBoost, followed by Spearman's correlation test. Six two-stage models were developed: Stage 1 stratified patients into high- and low-dose groups using a Dmax threshold (Dmax model), while Stage 2 further classified high-dose patients using a support vector machine (SVM) with selected RaVx features (RaVx-SVM model). Classification performance was assessed in training and test sets using bootstrap validation.
RESULTS: Grade ≥ 2 AOM occurred in 61.6 % of patients. The Dmax model with a 50 Gy(RBE) threshold achieved 87.2 % accuracy but had a high false-positive rate (FPR = 31.4 %). The RaVx-SVM model reduced false positives, with the RaV40 Gy(RBE)-SVM-model performing best. The integrated Dmax50&RaV40 Gy(RBE)-SVM model achieved 97.0 % accuracy (FPR 2.0 %) in training and 96.5 % mean accuracy (mean FPR 4.7 %) in the test.
CONCLUSIONS: Dose-volume radiomics effectively identified low-risk AOM patients in the high-dose group. The integrated model demonstrated high accuracy in identifying HNC patients with grade ≥ 2 AOM during CIRT.
METHODS AND MATERIALS: We retrospectively analyzed 190 HNC patients treated with CIRT (training: test = 7:3). Radiomic features were extracted from the entire oral mucosa and dose-specific subregions [RaVx, 10-50 Gy(RBE) in 10-Gy increments]. Feature selection employed logistic regression, random forest, and XGBoost, followed by Spearman's correlation test. Six two-stage models were developed: Stage 1 stratified patients into high- and low-dose groups using a Dmax threshold (Dmax model), while Stage 2 further classified high-dose patients using a support vector machine (SVM) with selected RaVx features (RaVx-SVM model). Classification performance was assessed in training and test sets using bootstrap validation.
RESULTS: Grade ≥ 2 AOM occurred in 61.6 % of patients. The Dmax model with a 50 Gy(RBE) threshold achieved 87.2 % accuracy but had a high false-positive rate (FPR = 31.4 %). The RaVx-SVM model reduced false positives, with the RaV40 Gy(RBE)-SVM-model performing best. The integrated Dmax50&RaV40 Gy(RBE)-SVM model achieved 97.0 % accuracy (FPR 2.0 %) in training and 96.5 % mean accuracy (mean FPR 4.7 %) in the test.
CONCLUSIONS: Dose-volume radiomics effectively identified low-risk AOM patients in the high-dose group. The integrated model demonstrated high accuracy in identifying HNC patients with grade ≥ 2 AOM during CIRT.