1. Prediction of Persistent Tumor Status in Nasopharyngeal Carcinoma Post-Radiotherapy-Related Treatment: A Machine Learning Approach.
- Author
-
Tseng, Hsien-Chun, Shen, Chao-Yu, Kao, Pan-Fu, Chuang, Chun-Yi, Yan, Da-Yi, Liao, Yi-Han, Lu, Xuan-Ping, Sheu, Ting-Jung, and Shen, Wei-Chih
- Subjects
PREDICTION models ,RECEIVER operating characteristic curves ,RADIOMICS ,POSITRON emission tomography computed tomography ,RETROSPECTIVE studies ,MEDICAL records ,ACQUISITION of data ,NASOPHARYNX cancer ,TREATMENT failure ,MACHINE learning - Abstract
Simple Summary: Radiotherapy is the primary and only curative treatment for nasopharyngeal carcinoma (NPC). The adoption of intensity-modulated radiotherapy (IMRT) alone or combined with optimized chemotherapeutic strategies has improved survival rates while reducing treatment-related toxicities. However, the persistent tumor status, including residual tumor presence and early recurrence, is the predominant cause of treatment failure, associated with poorer survival outcomes. This study extracted the radiomic features from pretreatment PET images, which were used to construct a prediction model to identify patients with NPC at high risk of having persistent tumors after treatment. The model trained by the AdaBoost algorithm showed significant diagnostic performance in predicting treatment failure. This model exhibited the feasibility of assisting clinicians in identifying high-risk patients before treatment, allowing for more personalized treatment plans and improving patient outcomes. Background/Objectives: The duration of the response to radiotherapy-related treatment is a critical prognostic indicator for patients with nasopharyngeal carcinoma (NPC). Persistent tumor status, including residual tumor presence and early recurrence, is associated with poorer survival outcomes. To address this, we developed a prediction model to identify patients at a high risk of persistent tumor status prior to initiating treatment. Methods: This retrospective study included 104 patients with NPC receiving radiotherapy-related treatment who had completed a 3-year follow-up period; 29 were classified into the persistent tumor status group and 75 into the disease-free group. Radiomic features were extracted from pretreatment positron emission tomography (PET) images and used to construct a prediction model by employing machine learning algorithms. The model's diagnostic performance was assessed using the area under the receiver operating characteristic curve (AUC), whereas SHapley Additive exPlanations (SHAP) analysis was conducted to determine the contribution of individual features to the model. Results: The prediction model developed using the AdaBoost algorithm and validated through five-fold cross-validation achieved the highest AUC of 0.934. Its sensitivity, specificity, positive predictive value, negative predictive value, and accuracy were 89.66%, 86.67%, 72.22%, 95.59%, and 87.5%, respectively. SHAP analysis revealed that the feature of high dependence low metabolic uptake emphasis
50 had the greatest impact on model predictions. Furthermore, patients classified as disease-free exhibited markedly higher overall survival rates compared with those with persistent tumor status. Conclusions: In conclusion, the proposed prediction model efficiently identified patients with NPC at a high risk of persistent tumor status by using radiomic features extracted from pretreatment PET images. [ABSTRACT FROM AUTHOR]- Published
- 2025
- Full Text
- View/download PDF