Author: David Brizel, Kyle J. Lafata, Jian-Guo Liu, Yvonne M Mowery, Yvonne M Mowery, William Paul Segars, Jack B Stevens 👨🔬
Affiliation: Department of Physics, Duke University, Carl E. Ravin Advanced Imaging Laboratories and Center for Virtual Imaging Trials, Duke University Medical Center, Duke University, Department of Radiation Oncology, Duke University, University of Pittsburgh 🌍
Purpose: To develop a technique to quantify tumor topology using a unifying mathematical framework that integrates texture and morphology and to evaluate its feasibility as a prognostic biomarker for early metabolic response of oropharyngeal cancer (OPC).
Methods: Briefly, our technique is as follows: First, we stochastically sample the primary tumor using a Langevin equation, where we consider CT and 18F-FDG-PET images as drift and drag forces, respectively. At each voxel location, we apply Langevin dynamics to sample the tumor’s topological landscape and explore interactions between structure and spatial heterogeneity of metabolism. Second, we calculate the mean free path, tau, of the Langevin trajectories, which represents the topological distance from a given voxel to the tumor boundary. Finally, the spatial encoding of tau values is used to create a topological map that intrinsically combines texture and morphology. Following numerical validation, the algorithm was applied as a proof-of-concept to 56 OPC patients enrolled on a prospective clinical trial to evaluate early treatment response to 20 Gy +/- concurrent chemotherapy. Features were extracted from the discretized tau distributions and compared with standard PET radiomics via cluster analysis and correlation matrices. Prognostic value was assessed using the Kaplan-Meier method, with patients stratified by median feature value and separation evaluated via log-rank tests (p<0.05).
Results: The algorithm showed strong agreement with analytical solutions (mean-absolute-error = 0.014). Distributions of tau values represented patient-specific tumor topology. Cluster analysis revealed that topological features provide information distinct from traditional PET radiomics, which was confirmed by variability in feature correlation between the two methods. Survival analysis demonstrated association between a representative topological feature and recurrence-free-survival (p = 0.013).
Conclusion: Our approach integrated different aspects of the tumor phenotype into a singular measure of tumor topology, and preliminary findings demonstrated its potential value in characterizing (chemo)radiation response in patients with OPC.