Exploring radiomic and topological features to evaluate treatment response in rectal cancer

Abstract Objective: This study investigates the potential of topological data analysis (TDA) in conjunction with conventional radiomics applied to longitudinal T2-weighted magnetic resonance imaging (MRI) to evaluate treatment response following neoadjuvant therapy, with or without anti-CD40 immunotherapy, in patients with locally advanced rectal cancer (LARC).
Approach: Paired pre- and post-treatment MRIs from 21 patients enrolled in the INNATE trial (NCT04130854) were analyzed, including Arm1 (radiation, chemotherapy, and anti-CD40 immunotherapy; n=12) and Arm2 (radiation and chemotherapy alone; n=9). Patients were categorized as having a maximal response (MR) or a partial response (PR). Three complementary delta-based analyses were performed: (1) 3D radiomics, (2) 2D radiomics and sliding-windows analysis, and (3) topological feature extraction via TDA. Statistical analysis included cross-arm comparisons (Arm1 vs. Arm2), response-based comparisons (MR vs. PR), and detailed pairwise subgroup analyses.
Main Results: For 3D radiomic features, ΔVolume effectively distinguishes between treatment strategies, with Arm1 exhibiting greater tumor reduction compared to Arm2. ΔElongation differentiates between response groups, with MR showing negative values indicative of increased tumor sphericity. In 2D radiomic analysis, Gray-level-dependence -matrix (GLDM) texture features computed across varying sliding windows differentiate between treatment arms but not response groups. Topological analysis identified six features derived from Betti numbers and persistence lifetimes, capturing local heterogeneity in tumor texture, shape and filtration patterns. Among these, five features significantly differentiate response groups, while one distinguishes treatment arms. At a finer level of subgroup comparison, ΔBetti₂ demonstrates significant differences, distinguishing MR-Arm1 vs. PR-Arm1 and MR-Arm1 vs. MR-Arm2.
Significance: This exploratory study suggests potential of topological features for outcome prediction in rectal cancer. These features may complement standard radiomics to enhance predictive performance as more data becomes available. Notably, the ability of topological features to distinguish maximal responders across treatment arms—with and without immunotherapy—may offer additional biological insights, potentially related to immune-mediated tumor response.