Author: Yongrui Bai, Xuming Chen, Yong Liu, Xiumei Ma 👨🔬
Affiliation: Department of Radiation Oncology, Renji Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China, Department of Radiation Oncology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China 🌍
Purpose: Hematologic toxicity (HT) is a common complication in patients with cervical or endometrial cancer. This study aims to develop a precise predictive model for acute HT in patients with cervical or endometrial cancer by integrating clinical risk factors, radiomics features, dosiomics features, and dose parameters.
Methods: A total of 207 patients with cervical or endometrial cancer from three cohorts who received pelvic external beam radiotherapy (EBRT) were retrospectively included in this study. We compared combinations of two normalization methods (Min-Max and Z-score), two feature selection methods (T-test + Lasso and mRMR), and six classifiers (SVM, SGD, KNN, RF, XGBoost, and LightGBM). Four submodels were developed: a clinical model, a radiomics model, a dosiomics model, and a dose parameter model. Additionally, the combination of variables from the best four submodels was used to construct an optimal integrated model. Given the severe class imbalance between negative and positive samples (167:40), model performance was evaluated using the area under the receiver operating characteristic curve (AUC).
Results: The best-performing clinical model, radiomics model, dosiomics model, dose parameter model and combined model achieved AUC values of 0.79 (95% CI, 0.71-0.86), 1.00 (0.98-1.00), 1.00 (0.91-1.00), 0.99 (0.98-1.00) and 0.91 (0.86-0.94), respectively, in the training cohort; 0.86 (0.71-0.95), 0.68 (0.52-0.82), 0.72 (0.52-0.85), 0.63 (0.50-0.77), 0.89 (0.72-0.95) respectively, in the internal test cohort; and 0.85 (0.71-0.94), 0.78 (0.63-0.89), 0.66 (0.51-0.80), 0.70 (0.55-0.83) and 0.88(0.74-0.96) respectively, in the external test cohort.
Conclusion: We proposed a combined predictive model for acute HT that integrates clinical risk factors, radiomics features, dosiomics features, and dose parameters. The results demonstrate the model outperforms the four individual submodels.
Funding: Natural Science Foundation of Zhejiang Province (No. LZ23A050002), National Natural Science Foundation of China (No. 12175012), the '111' center (B20065), the Fundamental Research Funds for the Central Universities, and a grant from Xiaomi Scholarship.