Predicting the Need for Adaptive Radiotherapy in Head and Neck Patients from CT-Based Radiomics and Pre-treatment Data

Abstract

Although adaptive radiotherapy can reduce the negative dosimetric and clinical impacts of anatomical changes during head and neck treatments, evidence shows that it is not equally beneficial for all patients. This makes it important to electively schedule adaptation ahead of time to optimize clinical resources and patient benefit. The purpose of this study is to assess the feasibility of using both pre-treatment patient features and radiomic features extracted from a pre-treatment contrast enhanced computed tomography scan to predict the need for adaptive radiotherapy. Seventy-two patients were included in the analysis, of which 36 required adaptation. 36 pre-treatment semantic features as well as 351 radiomic features extracted from the gross target volume were considered. Three support vector machine models were developed: 1) considering only semantic features; 2) considering only radiomic features; 3) using a combination of features from 1 and 2. A robustness analysis of the selected radiomic features was also conducted. The best classification results were obtained considering 6 features (4 semantic and 2 radiomic) with median accuracy and area under the receiver operating characteristic curve of 0.821 and 0.843, respectively.