ESTRO 2025 - Abstract Book

S3833

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Simulations were performed using a simplified two-compartment microstructural model 5,6 . For plausible ranges of DIL and NL model parameters 5 , ADCs were calculated for MRL and dMR acquisitions. Histology DIL was defined on whole-mount prostatectomy specimens stained with GLUT1. Within the DIL, stroma and tumour area fractions were calculated and correlated with ADC. Results: Ten patients were included: repeat MRL (n=2), repeat dMR (n=3), and one scan on each (n=5). Four were excluded: missing scan (n=2), image artefact (n=2); for one, DIL data only were excluded due to artefact. Intra-scanner repeatability (Figure 1a) was better for NL (wCV=1.7%) than DIL (wCV=4.2%), likely due to lesions’ relatively small size and typical location at the prostate edge. ADC tended to be higher on MRL than dMR (Figures 1b,c), hypothesised to be driven by the lower G max / b max . Simulated ADCs were in the same range as patient values, with inter-acquisition differences consistent with inter scanner differences in patients (Figures 1b,c).

DIL ADC tended to correlate positively/negatively with stroma/tumour area fractions, with MRL ADC yielding stronger correlations (Figure 2).