ESTRO 2024 - Abstract Book

S5118

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2024

2405

Proffered Paper

Voxel-based analysis of cognitive outcomes after radiotherapy for paediatric CNS tumours

Marianne C Aznar 1 , Lydia Wilson 2,3 , Angela Davey 1 , Eliana Vasquez Osorio 1 , Fakhriddin Pirlepesov 2 , Marcel van Herk 1 , Jason Ashford 4 , Heather Conklin 5 , Kate Vaughan 1 , Martin McCabe 1 , Thomas E Merchant 2 1 University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom. 2 St Jude Children's research Hospital, Radiation Oncology, Memphis, USA. 3 Jefferson University, Radiation Oncology, Philadephia, USA. 4 St Jude Children's research Hospital, Clinical Operations, Memphis, USA. 5 St Jude Children's research Hospital, Psychology and biobehavioral sciences, Memphis, USA

Purpose/Objective:

Advanced technologies can plan and deliver increasingly complex radiotherapy (RT) dose distributions, necessitating better knowledge of dose-response. Voxel-based analysis (VBA) methods aim to overcome limitations of conventional dose-volume-histogram (DVH)–based analyses by agnostically analyzing entire dose distributions (1). VBA methods have shown promise in the investigation of dose-outcome relationships across several body sites (e.g. lung, head and-neck and pelvis). It has been suggested that VBA is technically applicable to paediatric cohorts with brain/CNS tumours (2), but specific dose-outcome relationships have not yet been investigated in this patient population. In this study, we present the first results of VBA methods for predicting cognitive outcomes in childhood cancer survivors and compare VBA to traditional DVH analysis.

Material/Methods:

We used data from a published study of 124 children (aged 3-21 years) treated with craniospinal photon therapy for medulloblastoma (3) to 23.4 Gy and 36-39.6 Gy for average-risk and high-risk patients, respectively. All patients received a boost to tumour bed and chemotherapy (vincristine, cisplatin and cyclophosphamide). Outcomes included decline in working memory or processing speed, quantified via Reliable Change Index for cognitive assessments performed at baseline and annually up to 5 years post-treatment. Clinically meaningful cognitive decline was defined as a score below -1.645 (3). We investigated correlations with dose to the hippocampi (left and right), frontal white matter (left and right) and the corpus callosum (divided into genu, body and splenium). VBA comprised two steps: 1) spatial normalization of individual dose distributions to a reference anatomy and 2) statistical inference to evaluate associations between dose and cognitive outcomes. Our spatial normalization performed affine registration to roughly align individual patient images to the reference anatomy followed by deformable image registration to fine-tune the normalization using NiftyReg. The statistical inference used a permutation test of per-voxel t-test statistics to identify voxels in which dose was significantly associated with the presence of meaningful cognitive decline. Clinical variables (e.g. age and sex) were available but were not taken into account to allow comparison with the published DVH analysis.

We compared the significant volumes identified by VBA to the correlations between anatomic structures and outcomes evaluated in the prior study (in total, 12 combinations of structures and outcomes) where e.g., True Positive