Award Date

May 2023

Degree Type


Degree Name

Master of Science (MS)


Health Physics and Diagnostic Sciences

First Committee Member

Yu Kuang

Second Committee Member

Steen Madsen

Third Committee Member

Zaijing Sun

Fourth Committee Member

Shengjie Zhai

Number of Pages



The objective of this study was to develop a predictive model that utilizes a Light Gradient Boosting Machine (Light-GBM) to accurately assess the risk of severe radiation-induced toxicities in cancer patients undergoing chemoradiotherapy. The occurrence of these toxicities can have a substantial impact on the patients' quality of life. The Light-GBM model using clinical feature alone or integrating radiomic features extracted from computed tomography (CT) images used in treatment planning with clinical features can enhance the accuracy of early prediction.The study included 179 breast cancer patients and 223 patients with nasopharynx cancer who were treated with both radiotherapy and chemotherapy from April 2005 to October 2020. All of the patients with nasopharynx cancer had pre/postoperative CT/MR scans. The clinical features extracted from the medical records of patients included age, cancer stage, tumor size, tumor location, medical history, chemotherapy drugs, targeted therapy drugs, hormone therapy drugs, distant metastases, surgery, radiation therapy position, radiotherapy dose. The additional clinical features were also extracted from breast cancer patients, including electrocardiograph (ECG) signal score, and left ventricular ejection fraction (LVEF) value before radiotherapy. The Light-GBM was used to develop predictive models to predict oral mucositis and radiation dermatitis in nasopharynx patients and cardiotoxicity in breast cancer patients, respectively. The utility of the developed models was evaluated via receiver operating characteristic curve (ROC) and area under the curve (AUC). In breast cancer case, the patients were randomly divided into a training set (n=150) and a test set (n=29). A Light GBM enabled predictive model was developed using patients’ clinical features. The utility of the developed model was evaluated via ROC. The AUC of the cardiotoxicity Light-GBM model was 0.82, higher than the clinical significance threshold of 0.7. Age, LVEF value before radiotherapy, cancer position, targeted therapy, tumor stage, and hormone therapy were the most valuable influencing factors. Specifically, we found that more severe cardiotoxicity occurred in patients with older age, higher LVEF value before radiotherapy, later tumor stage, and abnormal ECG signals with bradycardia, tachycardia, T wave, and Q wave abnormalities. For nasopharyngeal cancer patients, we developed two Light-GBM machine learning models: Model A, which included only clinical features, and Model B, which combined radiomic and clinical features. The models were trained using a training set of 200 samples and validated with a test set of 20 samples. A total of 756 radiomic features were extracted from the planning target volume (PTV), gross tumor volume (GTV), clinical target volume-GTV (CTV-GTV), clinical target volume (CTV), and organs at risk regions in the images. The models' abilities to predict severe toxicities were evaluated using ROC analysis in the validation cohort. The AUC values for Model A, which predicted six different toxicities (radiation oral mucositis, radiation dermatitis, skin ulcer, sternocleidomastoid muscle toxicity, thyroid toxicity, and skin thickness toxicity), were 0.8, 0.71, 0.72, 0.68, 0.75, and 0.64, respectively. In contrast, Model B demonstrated increased AUC values of 0.86, 0.81, 0.84, 0.77, 0.89, and 0.8. Feature importance analysis revealed that T stage, age, radiation dose, chemotherapy drugs, and 14 radiomic features were the most valuable risk prediction factors. The results of this study illustrate the potential of utilizing machine learning models to predict various radiation-related toxicities. This approach facilitates the early identification of patients who may benefit from personalized chemoradiotherapy, timely interventions during or after chemoradiotherapy, or the use of alternative treatment technologies.


Medicine and Health Sciences

Degree Grantor

University of Nevada, Las Vegas




IN COPYRIGHT. For more information about this rights statement, please visit