Three-dimensional wavelet decomposition-based radiomics analysis for tumor characterization in patients with oropharyngeal squamous cell carcinoma

Abstract

Background: We investigated the potential predictive value along with interpretability of the three-dimensional wavelet decomposition (3D-WD)-based radiomics analysis for characterization of gross-tumor-volumes (GTVs) for patients with Human Papilloma Virus (HPV) oropharyngeal squamous cell carcinoma (OPSCC). The goal was to characterize and identify the spatial frequencies and regions of primary tumor that are responsible for classifying the HPV status. Methods: One-hundred twenty-eight OPSCC patients (60-HPV+ and 68-HPV-, confirmed by immunohistochemistry-P16-Protein) were retrospectively studied. 3D-WD analysis was performed on the contrast-enhanced-CT images of patients’ primary tumor-GTVs to decompose information into three decomposition levels explained by a series of high-pass and low-pass wavelet coefficients (WCs). Log-Energy-Entropy of the WCs was calculated as radiomics features. A Least-Absolute-Shrinkage-and-Selection-Operation (Lasso) technique combined with a Generalized-Linear-Model (Lasso-GLM) was applied on the feature space to identify and rank the frequency sub-bands associated with the HPV status. The classifier was validated using a nested-cross-validation technique. Average of Area Under ROC (AUC), and Positive and Negative Predictive values (PPV and NPV) were computed to estimate the generalization-error and performance of the classifier. The significant features were used to weight tumor sub-band frequencies to reconstruct the tumor zones with highest information towards characterization of HPV. Results: Among 22 frequency-based features, two low-frequency and two high-frequency features were statistically discriminant between the two cohorts. Results (AUC/PPV/NPV=0.798/0.745/0.823) imply that tumor’s high-frequency and low-frequency components are associated with its HPV positivity and negativity, respectively. Conclusions: This study suggests that compared to the central zones of tumor, peritumoral regions contain more information for characterization of the HPV-status. Albeit subject to confirmation in a larger cohort, this pilot study presents encouraging results in support of the role of frequency-based radiomics analysis towards characterization of tumor microenvironment in patients with OPSCC. By associating this information with tumor pathology, one can potentially link radiomics to underlying biological mechanisms.