Exploring the Role of CT-Based Delta-Radiomics in Unresectable Vulvar Cancer

<b>Background/Objectives:</b> To explore the prognostic potential of gross tumor volume (GTV)-based delta-radiomic features from CT simulation scans in patients with locally advanced unresectable vulvar cancer. <b>Methods:</b> A total of 21 patients (between 2019 and 2024) undergoing definitive radiotherapy were included, with baseline and post-phase I (after 25 fractions) CT simulation scans analyzed. Radiomic features (<i>n</i> = 107) were extracted from GTVs using PyRadiomics, and delta features were calculated as the relative change between scans. A multi-step selection pipeline (univariable Cox screening (<i>p</i> < 0.10), correlation filtering, and Lasso-Cox) was applied for each endpoint: local control (LC), regional control, distant metastasis-free survival, progression-free survival, and overall survival (OS). Model discrimination was assessed via 500-iteration bootstrapped concordance index (C-index), and calibration was plotted at 24 months. <b>Results:</b> Median follow-up was 50.0 months. The 2-year LC and OS rates were 56.2% and 55.9%, respectively. Final multivariable models retained a sole texture Δ feature for LC (HR = 2.62, 95% CI = 1.05-6.52, <i>p</i> = 0.039; C-index = 0.748) and six Δ features for OS (C-index = 0.864). No features were retained for other endpoints. For LC, increased run-length non-uniformity after phase I predicted poorer control. For OS, increased texture/shape complexity predicted worse survival, whereas increased uniformity predicted better survival. <b>Conclusions:</b> CT-based delta-radiomic features, particularly shape and texture metrics, may predict LC and OS in unresectable vulvar cancer. Despite the small sample size, these findings highlight the potential for delta-radiomics as a noninvasive biomarker for risk stratification. Validation in larger cohorts and exploring potential in adaptive radiotherapy are warranted.