A Radiomics-Based Classifier for the Progression of Oropharyngeal Cancer Treated with Definitive Radiotherapy.

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Oropharyngeal cancer (OPC) is often HPV-positive.
• Radiotherapy is a standard treatment for OPC.
• Predicting disease progression is crucial for treatment individualization.

Purpose of the Study:

• To evaluate radiomics features from pre-treatment PET scans for predicting disease progression in HPV-positive OPC.
• To compare machine learning model performance using different feature selection methods.

Main Methods:

• Retrospective analysis of PET images from two Mayo Clinic cohorts (n=72 training, n=22 testing).
• Radiomics and clinical features were extracted and analyzed using Mann-Whitney U test.
• Machine learning models (Gaussian Naïve Bayes) were developed and validated using forward feature selection.

Main Results:

• Machine-driven radiomics features demonstrated superior performance and reduced overfitting compared to manually filtered features.
• A four-variable model incorporating 'Radiation Type' and three radiomics features achieved 79% training and 77% testing accuracy.
• Radiomics features provided risk stratification beyond HPV status.

Conclusions:

• Pre-treatment PET radiomics can predict disease progression in HPV-positive OPC patients.
• Machine learning models utilizing radiomics enhance risk stratification for personalized treatment and follow-up.