Author: Giacomo Avesani, Matthew D Blackledge, Silvia Bottazzi, Alison Tree, Nina Tunariu 👨🔬
Affiliation: Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Institute of Cancer Research, The Royal Marsden NHS Foundation Trust 🌍
Purpose: To evaluate changes in median apparent diffusion coefficient (ADC) within metastatic oligoprogressing lesions from prostate cancer treated with stereotactic body radiotherapy (SBRT) and compare these changes with untreated control lesions.
Methods: Twenty-eight patients with oligoprogressing bone disease underwent SBRT to up to two target regions. Whole-body MRI was performed before and 134–297 days post-SBRT using a standardized protocol with three b-values (50/600/900 s/mm2). Treated and up to two untreated lesions per patient were delineated by an expert radiologist using all available imaging. A separate cohort of 10 metastatic prostate cancer patients underwent pre-treatment test/retest baseline scans on the same day using the same protocol; up to 10 distinct lesions per patient were delineated. Novel Bayesian mixture modelling was applied to median tumor ADC values, incorporating the test/retest cohort to estimate measurement repeatability, and provided estimates of (i) μΔ, the population-level median ADC change for significantly responding lesions, and (ii) λ, the proportion of lesions demonstrating significant changes (range: 0 to 1).
Results: Forty-two treated and thirteen control lesions were identified in SBRT-treated patients, with seventy-three lesions delineated in the test-retest dataset. Treated lesions showed a significant median ADC increase (p < 0.001), unlike controls (p = 0.10). Bayesian mixture modelling confirmed this, estimating μΔ = 0.85×10-3 mm2/s (95% confidence interval: 0.72–0.97) for treated lesions, while controls showed an insignificant population ADC change of μΔ = 0.21×10-3 mm2/s (CI: -1.24–2.46, wide CI due to low numbers in this subgroup). The proportion of significant changes at the lesion level was higher in treated lesions: differnece in λ = +0.52 (CI: 0.01–0.83).
Conclusion: ADC is a sensitive biomarker for detecting inter-tumoral response heterogeneity and monitoring treatment success. Our Bayesian mixture model offers a robust framework for simultaneously estimating population parameters and identifying non-responding lesions within an individual.