ESTRO 2024 - Abstract Book

S2046

Clinical - Paediatric

ESTRO 2024

Developmental Biology, Manchester, United Kingdom. 4 St Jude Children's, Department of Radiation Oncology, Memphis, USA

Purpose/Objective:

Growth hormone deficiency (GHD) is a common side effect for children treated with radiotherapy for brain tumours. If GHD is left untreated, it can impact quality of life with, e.g., permanent short stature and impaired health. Dose to the hypothalamus-pituitary axis has previously been identified as a contributing factor in GHD, but brain substructures have not been considered independently, and dose-response is poorly understood [1]. To address this, we utilise the voxel-based analysis technique known as image-based data mining (IBDM) to explore dose-response with no a-priori hypothesis about which brain substructures are linked to the outcome [2]. In our study, we present the first application of IBDM to study GHD in two independent cohorts of childhood cancer survivors.

Material/Methods:

Data was available for 261 children treated with photon radiotherapy for brain tumours across two clinical centres. We analysed this data using our IBDM toolkit, which has been validated for use in paediatrics [3]. IBDM includes spatial normalisation of dose distributions to a reference anatomy and voxel-wise analysis of dose and outcome. For centre A, planning T1 MRIs were collected for 146 patients with medulloblastoma. The cohort for centre B included 115 patients with mixed diagnoses, for which planning CTs were collected. The choice of CT or MRI has been shown to have little impact on the IBDM process in this patient population [4]. A reference patient of median brain volume was selected for each cohort independently. Following spatial normalisation (affine followed by deformable registration using NiftyReg ), each patient’s dose distribution was blurred to account for inaccuracy in registration [5]. For statistical analysis, a per voxel t-test statistic was used to represent differences in dose between patients with differing GHD status (yes vs no). To highlight voxels with a significant association between dose and GHD, permutation testing (n=1000) was applied using the maximum t-statistic. For anatomic interpretation, FreeSurfer was used to segment brain substructures on both reference patients. The volumetric overlap between the IBDM identified region and labelled substructures was calculated. For each cohort, the mean dose to the 99% significance region was calculated and included in a multivariable logistic regression adjusting for clinical variables. Results were compared across the two independent analyses to validate whether the same dose-response relationships were observed.

Results:

Available data is shown in Table 1.

Centre A, N = 146

Centre B, N = 115

Growth hormone deficiency (GHD)

Yes

114 (79%)

46 (42%)

No Unknown

30 (21%) 2

64 (58%) 4

Diagnosis