ESTRO 2024 - Abstract Book

S5031

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

Purpose/Objective:

Various methods have been proposed in the literature for studying the relationship between toxicity outcomes and dose in radiotherapy. Dose constraints based on dose-volume histograms (DVH) are commonly employed to strike a balance between effectively treating the tumour while preserving surrounding healthy organs. However, the major limitation of DVH analysis is the absence of spatial dose information in organs at risk (OARs). Voxel-based techniques, such as dose-response maps (DRM) derived from dose-surface maps (DSM), aim to identify high-risk sub-regions in OARs. The optimal selection of parameters, such as coordinate systems and statistical approaches, remains a subject of debate, potentially affecting the output of experiments and their potential applicability in clinical practice. This study aims to assess the sensitivity of several parameters in identifying high-risk sub-regions on DRM of the rectum and bladder in prostate cancer (PCa) patients undergoing radiotherapy.

Material/Methods:

We analysed data from 1808 PCa patients (who underwent radical radiotherapy and post-prostatectomy who received salvage/adjuvant radiotherapy) participating in the REQUITE trial. REQUITE stands for validating pREdictive models and biomarkers of radiotherapy toxicity to reduce side effects and improve QUalITy of lifE in cancer survivors. REQUITE is a prospective multicentre cohort study conducted across 26 hospitals in 8 countries from April 2014 to March 2017, with the primary objective of validating predictive models or biomarkers for assessing the likelihood of post-radiotherapy toxicities in cancer patients. The rectum and bladder were segmented on planning CT scans using ADMIRE software (ELEKTA, Sweden). Rectum DSMs were spatially normalised using eight cylindrical coordinates: fixed extent 22-40mm inferior and superior (combined) from the prostate centre, 30mm inferior or 30mm superior (separately) to the prostate centre, and a 91x90 fixed-sized grid covering the entire length of the rectum. Bladder DSMs were spatially normalised using two methods: cylindrical and spherical coordinates with a 91x90 fixed-sized voxel grid. Doses were converted to equivalent doses in 2 Gy fractions (EQD2) with α/β ratios of 1, 2, 3, 4, and 5 Gy. We assessed 27 clinician-reported late toxicities through Clinician-rated Common Terminology Criteria for Adverse Events (CTCAE) scores and 23 patient-reported outcomes (PROs) based on a pelvic symptoms questionnaire. Late toxicities were normalised by subtracting the baseline toxicity value from the maximum of 12- and 24-month toxicity values, and dichotomised at grades 1 or 2. We performed a binary analysis of DSMs based on dichotomised toxicity values. We analysed DSMs voxel-wise using Welch's t-test and conducted permutation testing to correct for multiple comparisons, with 10, 100, and 1000 iterations. DRMs were derived from the DSMs by calculating the test distribution (Tmax) as the maximum (over the whole image) of the difference between the means for outcome groups dividing by the overall standard deviation, and thresholding at the 90th and 95th percentiles to identify statistically significant voxels associated with specific toxicities.

Results:

Visual comparison showed a range of differences in the identified high-risk sub-regions with respect to changes in the parameter’s values. The DRM obtained using spherical coordinates varied less than those obtained using cylindrical coordinates (Figure 1) for the bladder with changing permutation numbers (Figure 2), suggesting that unwrapping it using a spherical coordinate system may be more appropriate. Image extent and grade threshold changed the number of toxicity events in each analysis, directly affecting the identified location and the number of significant voxels. With more events typically larger regions were found. When the number of events was fixed and the image extent was varied, or when the image extent was fixed and the number of events was varied (within the same grade by changing the number of datasets included in the analysis), a range of differences (from small to large,