Author: Todd A Aguilera, Gaurav Khatri, Jiaqi Liu, Hao Peng, Nina N. Sanford, Robert Timmerman, Haozhao Zhang 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, UT southwestern medical center, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center 🌍
Purpose:
This study first integrates 3D topological data analysis with radiomics from local advanced rectal cancer T2-weighted MRI to evaluate therapeutic responses and quantify treatment-induced changes following NCRT ± anti-CD40 immunotherapy (NCT04130854), aiming to identify novel biomarkers for treatment stratification.
Methods:
Pre- and post-treatment MRIs from 21 INNATE trial patients (Arm1: NCRT + anti-CD40, n=12; Arm2: NCRT alone, n=9) were analyzed. Gross tumor volumes (GTVs) were normalized to [0,1], and sublevel set filtrations constructed persistent homology representations. Topological features such as Betti numbers (H₀, H₁, H₂) and derived metrics (e.g., persistence entropy, lifespan statistics) quantified spatial heterogeneity and treatment-induced changes. Gaussian modeling of Betti₂ dynamics employed fitted parameters (a, µ, σ, R²) as secondary features. Comparative analyses were conducted to evaluate the capability of anatomy-registered 2D axial radiomic features, extracted from pre- and post-treatment slices, and 3D whole-GTV radiomic features in stratifying complete (CR) and partial response (PR) groups.
Results:
Arm1 exhibited greater GTV reductions (∆Volume: -47.64±30.24 cm³ vs. Arm2: -18.56±22.44 cm³, p<0.0095), yet conventional radiomics showed limited discriminatory power. In contrast, topological features revealed significant pre- and post-treatment differences. CR_Arm1 showed elevated pretreatment H₂ cavities (Betti₂ at t=0.2; 53.75±23.77 vs. other groups, p<0.05) and significant post-treatment reductions (Δ=-143.0 ± 138.72 vs. Δ=0.53 ± 10.13 in others, p<0.0002), uncorrelated with GTV changes (Spearman ρ=0.35). Gaussian modeling distinguished responders (75% CR with R² < 0.95 pre-treatment) from non-responders (77% PR with R² ≥ 0.95). In PR_Arm2 (non-responders without anti-CD40), minimal delta changes in persistence entropy (dim2) (Δ=-0.52 ± 0.44 vs. Δ≤-2.31 in others, p<0.0005) indicated limited structural remodeling, consistent with poor outcomes.
Conclusion:
3D topological features outperformed conventional radiomics by providing biologically interpretable insights into immune-specific spatial remodeling and therapeutic response. These findings highlight the potential of anti-CD40 to enhance treatment evaluation and support integrating topology-based biomarkers into clinical workflows.