ESTRO 2025 - Abstract Book

S3720

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Purpose/Objective: The objective of this study was to develop and validate a method for computing K trans maps within a research version of a treatment planning system (TPS) to enable accurate, repeatable, and reproducible estimation of tumor blood flow and vascular permeability 1 .

This method was validated using the OSIPI DCE-MRI challenge framework 2 to ensure consistency and reliability compared to existing in-house, often more manual solutions.

Material/Methods: The implementation uses the extended Tofts model 3 and voxel-wise, linear fitting. A candidate, patient-specific vascular input function (VIF) is automatically detected from the superior sagittal sinus (SSS). Input images are preprocessed using Gaussian filtering. The challenge framework 2 consists of scripts and a dataset containing test-retest scans of 8 patients with glioblastoma 4 for assessing repeatability, and to evaluate accuracy, two pairs of synthetic scans. Reproducibility is investigated by having two teams independently analyze the data following a standard operating procedure (SOP) developed for the implementation. Results: Computing maps takes ~1 minute per DCE-MRI series. Compared against previous challengers 2 , we rank first in repeatability and reproducibility, and second in accuracy (see Table 1, TPS(n=3)). Two factors harming accuracy were identified: 1. By default, the TPS implementation computes the pre-contrast signal by averaging n=3 initial time points. Increasing n, while still only including pre-bolus images, improves results (Table 1, TPS(n=7)). 2. The synthetic data has an implausible signal in the SSS, resulting in a poor candidate VIF detection (Fig.1). Therefore, based on an automatic warning from the software, both test teams chose to replace the candidate with a software-provided population-based alternative 5 according to the SOP. Table 1 : Results for TPS and the top two previous challengers based on OSIPI gold 2 . For n=7, reproducibility is marked "?", since only one group ran this experiment (for n=3, both groups performed the analysis). We expect reproducibility to be independent of n, since the implementation is deterministic and the only manual input is determined by the SOP (candidate VIF accepted, unless the system suggests using Parker 5 ).