Author: Waleed Mutlaq Almutairi, Ke Colin Huang, Vishwas Mukundan, Christopher F. Njeh, Oluwaseyi Oderinde, Yong Yue 👨🔬
Affiliation: Purdue University, Indiana University School of Medicine, Department of Radiation Oncology, Advanced Molecular Imaging in Radiotherapy (AdMIRe) Research Laboratory, Purdue University, West Lafayette, Indiana, USA 🌍
Purpose:
This study aimed to develop a machine learning (ML) model for early prediction of chemoradiotherapy (CRT) response in order to enhance personalized treatment selection for oral or oropharyngeal cancer (OPC) patients.
Methods:
This retrospective study selected 42 stage I and II OPC patients from the Cancer Imaging Archive (QIN-HEADNECK) who underwent CRT. Pre-treatment FDG-PET/CT images for the patients were retrieved, and lesion volumes were delineated using a 20% SUVmax. Radiomic features were extracted from PET and CT images, and 15 features were selected based on the performance of Variance Inflation Factor (VIF) and Minimum Redundancy Maximum Relevance (MRMR) analysis. Machine learning models were developed using the Logistic Regression (LR) and Random Forest (RF) algorithms, and the performance of the models was assessed using area under the curve of receiver characteristics (ROC-AUC) and accuracy.
Results:
The RF model achieved an overall AUC of 0.88 (95% CI: 0.7638 - 0.9593), accuracy of 80.7%, sensitivity of 75%, and specificity of 86.7%. Comparatively, the LR achieved an overall AUC of 0.83 (95% CI: 0.7081 - 0.9454) and accuracy of 80.7%, with a sensitivity of 86% and a specificity of 83%.
Conclusion:
The RF model demonstrated a statistically significant performance in identifying early CRT response, with the potential to improve treatment outcomes. Integrating this ML model into clinical workflows could support clinical decision-making to select OPC patients for CRT.