ESTRO 2023 - Abstract Book

S644

Monday 15 May 2023

ESTRO 2023

Conclusion A B-LSTM can achieve state-of-the-art prediction accuracy with the added benefit of comprehensively quantifying the magnitude and spatial distribution of the prediction uncertainty. Even with ensembles of 100 predictions, the runtime overhead stayed below a factor of 10 and could be further reduced by pruning and computational optimization. Expanding the patient training dataset may improve prediction accuracy and the Bayesian approach may similarly be applied to other recent approaches (e.g., transformer networks as used in Pastor-Serrano & Perko 2022). Thus, B-LSTMs may serve as a first step towards comprehensible deep-learning dose prediction, enabling clinical translation in the future. OC-0776 Are MRI scans needed for precise image-based data mining in paediatric brain tumour cohorts? A. Bryce-Atkinson 1 , E. Vasquez Osorio 1 , A. Green 2 , A.M. Faught 3 , M.G. McCabe 1 , T.E. Merchant 3 , M. van Herk 1 , M.C. Aznar 1 , L.J. Wilson 3 1 The University of Manchester, Division of Cancer Sciences, Faculty of Biology, Medicine and Health, Manchester, United Kingdom; 2 European Bioinformatics Institute, European Molecular Biology Laboratory, Hinxton, United Kingdom; 3 St Jude Children's Research Hospital, Department of Radiation Oncology, Memphis, USA Purpose or Objective Image-based data mining (IBDM) is a voxel-wise technique for identifying spatial dose-response associations in radiotherapy. IBDM spatially normalises 3D imaging and dose distributions to a reference anatomy prior to computing ‘per-voxel’ correlations of dose and clinical outcomes. Applying IBDM in paediatric cancer cohorts often requires multi-institutional collaborations to achieve appropriate cohort s for analysing long-term retrospective data. A particular challenge for IBDM brain-tumour research is that although MRI is indicated due to superior anatomical definition, it may not be as readily available as CT in historical cohorts. This work compares results from MRI- and CT-based IBDM pipelines to evaluate the impact of using MRI for IBDM paediatric studies in the brain. Materials and Methods We used CT, MRI, radiotherapy dose and contours from 128 children who received curative craniospinal irradiation for Medulloblastoma. Spatial normalisation was performed by non-rigidly registering each patient’s scan to the patient with the median brain volume. We quantified spatial normalisation accuracy by comparing each patient’s mapped structure contours to the corresponding reference contours via mean distance to agreement (mDTA). To standardize the dose comparison, we created modified dose distributions by replacing the dose in 5 structures (Figure 1) with the mean dose in that structure and simulated outcome labels, assigning “effect” labels to patients whose mean structure dose exceeded the median dose of the cohort. Per-voxel dose comparisons used permutation testing (1000 permutations) to identify voxel clusters in which the dose was significantly associated with the simulated outcome labels. Identified significant clusters were compared between the CT- and MRI-based pipelines via mDTA. In this test, the “ideal” IBDM result would be significant clusters of voxels that match the structure containing the dose modification. Proffered Papers: Novel applications of MR imaging

Results