University of Maryland Radiation Oncology Baltimore, MD, United States
X. Ling1, S. Bazyar2, M. Ferris2, J. K. Molitoris2, E. Allor1, H. Thomas1, D. Arons1, L. Schumaker1, R. F. Krc3, W. Mendes2, P. T. Tran4, R. Mehra5, D. Gaykalova6, and L. Ren2; 1University of Maryland School of Medicine, Baltimore, MD, 2Department of Radiation Oncology, University of Maryland School of Medicine, Baltimore, MD, 3Department of Radiation Oncology, University of Maryland Medical Center, Baltimore, MD, 4Department of Radiation Oncology, Division of Translational Radiation Sciences, University of Maryland Baltimore, School of Medicine,, Baltimore, MD, 5University of Maryland Cancer Center, Baltimore, MD, 6University of Maryland School of Medicine, BALTIMORE, MD
Purpose/Objective(s): This study aims to identify imaging biomarkers associated with progression-free survival (PFS) for advancing precision medicine and improving outcomes in customized treatment optimization to increase the survival rate for head and neck squamous cell carcinoma (HNSCC) patients. Materials/
Methods: Computed tomography scans were collected from 147 treatment-naïve patients with HNSCC. We extracted 1,092 radiomic features from gross tumor and gross nodal volumes in each patient’s pre-treatment CT, encompassing both initial diagnoses and recurrent cases. Clinical characteristics were also obtained, including race, sex, age, tobacco and alcohol use, tumor staging, and treatment modality. An ad hoc feature selection algorithm was used to eliminate the most redundant features, followed by a selection of the best subset of the Cox proportional hazard model (CPH). The final CPH was chosen among significant models (likelihood ratio and score test both significant with p < 0.05) and contained at least one significant variable (Wald test, p < 0.05). The best CPH model is determined based on the best prediction accuracy in terms of the concordance index. Similarly, significant input radiomic features in the final CPH model were identified as imaging biomarkers. The model was evaluated through a 10-fold cross-validation. Results: The final results of the CPH model show that the wavelet-HLH filtered first-order-maximum (FOM) has a Wald statistic of -2.14 and a p-value of 0.03, with a coefficient of -0.99. This suggests that a tumor with a one-unit increase in FOM is 30 times less likely to progress compared to a tumor without the increase. The group with FOM > 2.99 showed a 3-year progression-free survival rate of 0.81, compared to 0.49 for the group with FOM = 2.99. The relative risk for smokers was found to be 1.9, which means that non-smokers have a lower risk of progression compared to smokers. Conclusion: Our approach identified significant CT-based imaging biomarkers for PFS in patients with HNSCC. Incorporating radiomic analysis into clinical practice could improve decision support by stratifying patients for treatment customization to improve outcomes.
