ESTRO 2025 - Abstract Book

S59

Invited Speaker

ESTRO 2025

inherent in MRI data collection and subsequent analysis. General recommendations and guidelines for estimating and reducing these variations are presented, emphasizing the need for standardized procedures to ensure reproducibility. Special attention is given to the robustness of diffusion-based metrics, with illustrative examples focusing on the extraction of the apparent diffusion coefficient (ADC) and intravoxel incoherent motion (IVIM) parameters. The talk will provide examples on how careful optimization and standardization of imaging protocols can significantly reduce variability in these parameters, fx. through the use of optimal b-values. In addition, alternative strategies based on model-free parameter extraction will be addressed as a means to further mitigate potential biases introduced by model-dependent analyses. As an example of model-free analyses, the recent monotonous slope non-negative matrix factorization (msNMF) will be discussed. Lack of specificity in conventional diffusion metrics will also be discussed providing examples from recent work in so-called time dependent diffusion MRI, which enables better disentanglement of confounding microstructural features of the tissue. By comparing examples from both ADC/IVIM, model-free approaches, and advanced MRI techniques, this session aims to provide practical insights into the development of robust quantitative MRI protocols, ultimately facilitating their integration into large-scale, multicentre trials. The overarching goal is to enhance the reliability of MRI biomarkers for radiotherapy response assessment, providing more effective clinical decision-making and improved patient outcomes in future patients.

4742

Speaker Abstracts Challenges and opportunities of deep learning quantitative MRI parameter estimation Andreas Wetscherek Joint Department of Physics, The Institute of Cancer Research, London, United Kingdom

Abstract:

Opportunities:

Deep learning (DL) could be applied at many different stages of the quantitative MRI pathway 1,2 :

1. Image Acquisition 2. Image Reconstruction 3. Image Postprocessing 4. Quantitative Analysis 5. Image Segmentation 6. Feature Extraction 7. Clinical Decision Making

In quantitative MRI, we often acquire a series of MRI scans, varying one or two acquisition parameters to enable calculation of quantitative tissue parameters. Each additional scan prolongs the total acquisition time, thereby decreasing the likelihood of clinical acceptance, patient tolerance and increasing the chance of motion-related artefacts that require correction. DL can help identify the most relevant parameter settings and enable quantitative parameter estimation from fewer scans. Furthermore, the acquisition time of individual scans can be reduced by using DL for image reconstruction from undersampled data or by applying DL to enhance images. Examples are denoising and increasing the reconstructed resolution beyond the acquired resolution (super-resolution). Typical