ESTRO 2025 - Abstract Book

S3791

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

2564

Digital Poster Texture Feature Stability and Reproducibility for Assessment of Early Radiotherapy Response in Uterine Cervical Cancer Ulrika Björeland, Kristina Sandgren, Sara Strandberg Department of Diagnostics and Intervention, Umeå University, Umeå, Sweden Purpose/Objective: In uterine cervical cancer (UCC), early radiotherapy (RT) response assessment with magnetic resonance imaging (MRI) may be enhanced using textural features (TF)[1-3]. TF reproducibility and stability can differ between MRI scanners and sequences [4]. Therefore, this study aims to assess TF stability, reproducibility, and reliability for detecting early structural changes in MRI in UCC after one week of RT. Material/Methods: The first 11 UCC patients from the prospective PRODIGYN study (NCT05855941) were included. TF within regions of interest (ROIs) were evaluated in T2-weighted (T2W) MRI performed at baseline and after one week of RT. For each ROI, 165 TF was obtained with SPAARC software [5] following biomarker guidelines [6] and nine first-order statistics features. In baseline T2W, two ROIs were placed in the right and left femoral head. The T2Ws were rigidly registered on the bone structures to ensure evaluation of the same regions in the two series. To assess reproducibility, the Wilcoxon signed-rank test compared the TF between the ROIs. Parameters without significant differences were considered stable. Intraclass Correlation Coefficient (ICC) analysis was performed on normalized TF to assess the consistency of these stable parameters between the ROIs. The last step was calculating the correlation matrix of stable features for clustering analysis to visualize the relationships between features. All tests were performed in R software. The identified stable features were applied to a subsample of five UCC patients to assess the potential feasibility of monitoring early tissue changes during the first week after RT. Results: The Wilcoxon test showed that 141 TF did not differ significantly between the two T2Ws (p>0.05), indicating high reproducibility and stability. Furthermore, 29 TF displayed a significant difference, and 5 TF were consistent. ICC analysis on the reproducible TF showed that only 7 texture indexes exhibited strong consistency (ICC > 0.8). Cluster analysis revealed the correlation between features, and four representative, stable features were selected (table 1). For the subsample of five UCC patients, the change from baseline to one week of RT was visualized (figure 1) for different ROIs, with a difference in response patterns in tumour ROIs and organs at risk (rectal and bladder walls).