OS03.5.A Corelation between longitudinal t2 MRI radiomic primary texture feature values and radiation dose in non-tumoral regions of the brain in paediatric brain tumours

• Published in: Neuro-oncology (Charlottesville, Va.) Va.), 2021-09, Vol.23 (Supplement_2), p.ii6-ii6
• Main Authors: Sakhavalkar, P, Avula, S, Pizer, B, Thorp, N, Jenkinson, M
• Format: Article
• Language: English
• Subjects: Oral Presentations
• ISSN: 1522-8517; 1523-5866
• DOI: 10.1093/neuonc/noab180.018

Abstract BACKGROUND Paediatric brain tumour survivors may have treatment toxicity associated with signal change on follow-up MRI. Quantitative MRI texture features can potentially be used as surrogates of the underlying tissue changes following radiation therapy. MATERIAL AND METHODS Longitudinal retrospective study in 51 paediatric primary brain tumours treated with photon (N=30) and proton (N=21) radiotherapy (RT). T2 MRI scans at baseline and multiple time point from the date of surgery to 2 years following radiotherapy were selected for the textural analysis. Scans were bias corrected, registered with the CT dose maps and with baseline scan for each patient using 3Dslicer. Regions of interest (ROI) of fixed diameter were drawn in 11 predetermined non-tumoral regions of brain including in peri-tumoural region (PTV). ROIs were placed in homogenous white/grey matter. Radiation dose was calculated in each of these 11 ROIs and texture features were extracted using pyradiomics. Data were analysed using machine learning and statistical analysis. General linear multivariate model was used to corelate primary texture features over period of 24 months and radiation dose, time, effect of dose*time together at each ROI separately. RESULTS There were Brainstem 4, Cerebellar19, Hemispheric cerebral 7 and Supratentorial midline 10 tumours. Median age at diagnosis was 8.26 years (range: 0–20). Median RT treatment dose was 28.52Gy (0-60Gy). Multivariate analysis shows significant corelation (p < 0.001) between radiation dose and longitudinal primary texture features in all 11ROIs. Time showed corelation with feature values only in 3 ROIs and dose* time showed corelation in 5ROIs. Primary (statistical) feature values showing consistent correlation with dose in all 11 ROIs over 24 months are total energy, 10%, 90%, energy, entropy, mean, median, and minimum. CONCLUSION Radiomic texture analysis is a promising modality to understand dose related textural changes in the normal part of brain in paediatric brain tumour patients treated with radiation therapy. Radiomic changes need to be related to neurological outcomes in future research.