AUTHOR=Wu Kuan , Miu Xiaoyan , Wang Hui , Li Xiadong TITLE=A Bayesian optimization tunning integrated multi-stacking classifier framework for the prediction of radiodermatitis from 4D-CT of patients underwent breast cancer radiotherapy JOURNAL=Frontiers in Oncology VOLUME=Volume 13 - 2023 YEAR=2023 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2023.1152020 DOI=10.3389/fonc.2023.1152020 ISSN=2234-943X ABSTRACT=Purpose: we aimed to develop a Bayesian optimization based multi-stacking deep learning platform for prediction of radiation-induced dermatitis (grade ≥ two) (RD 2+) before radiotherapy by using multi-region dose-gradient-related radiomics features on base of pre-treatment four-dimensional planning computed tomography (4DCT) images, clinical and dosimetric data from patients under went Radiotherapy. Materials and methods: 214 patients with pathological diagnosis of breast cancer underwent radiotherapy after breast surgeries were retrospectively admission to the research cohort. 3 PTV dose related ROIs, including irradiation volumes covered by 100% (PTV_100PD), 105% (PTV_105PD), 108% (PTV_108PD) combined with 3 skin dose related ROIs, including SKIN_20Gy, SKIN_30Gy and SKIN_40Gy, were delineated, in which a total of 4309 features including radiomics features combined with clinical and dosimetric characteristics were extracted, 9 mainstream deep machine learning algorithms and 3 stacking classifier algorithm were employed for modeling training and validation. A Bayesian optimization based multi-parameter tuning technology was adopted for AdaBoost, RF, DT, GB and XTree machine learning models. The 5 parameter tunned learners and the other 4 learners ('LR', 'KNN', 'LDA', 'Bagging') whose parameter that cannot be tuned all as the primary week learners are finally sent to the subsequent meta-learner for subsequent training and learning. Results: 20 radiomics features, 8 clinical and dosimetric variables were selected for model training and validation, At the primary learner level, on base of Bayesian parameter tuning optimization, the random forest, XGBoost, AdaBoost, GBDT, and LGBM models with the best parameter combinations achieved AUC of 0.82, 0.82, 0.77, 0.80, and 0.80 prediction performance in the verification data set, respectively. In the secondary meta-learner lever, compared with LR and MLP meta-learner, the best predictor of symptomatic RD 2+ for stacked classifiers was the GB meta-learner with an area under the curve (AUC) of 0.97 [95% CI: 0.91-1.0] and an AUC of 0.93 [95% CI: 0.87-0.97] in the training and validation datasets, respectively and the 10 top predictive characteristics were identified. Conclusion: A multi-region dose-gradient-based Bayesian optimization tunning integrated multi-stacking classifier framework can achieve a high-accuracy prediction of symptomatic RD 2+ in breast cancer patients than any other single deep machine learning algorithm. cking Classifier