Distinguishing lymph nodes in head and neck cancer patients using MRI-based radiomics

The identification of the primary tumor location in patients with head and neck carcinoma of unknown primary involves an invasive and complex diagnostic protocol, thus fostering the development of non-invasive methods. Herein, a radiomic-based approach was proposed to distinguish between oropharynx and nasopharynx primary tumor location from the lymph nodes segmented in magnetic resonance images of head and neck cancer (HNC) patients. A total of 200 HNC patients (100 oropharynx and 100 nasopharynx) were considered. 10-fold cross-validation with class proportion was applied. Five different feature selection methods and five machine learning classification algorithms were tested. Overall, a high classification performance was achieved by all the combined feature selections/machine learning algorithms, with the best results obtained from the support vector machine and the neural networks algorithms with neighborhood component analysis (acc 100%).