Radiological Predictors of Cognitive Impairment in Paediatric Brain Tumours Using Multiparametric Magnetic Resonance Imaging: A Review of Current Practice, Challenges and Future Directions

Abstract

Paediatric brain tumours and their treatments are associated with long-term cognitive impairment. While the aetiology of cognitive impairment is complex and multifactorial, multiparametric Magnetic Resonance Imaging (MRI) can identify many risk factors including tumour location, damage to eloquent structures and tumour phenotype. Hydrocephalus and raised intracranial pressure can be observed, along with risk factors for post-operative paediatric cerebellar mutism syndrome or epilepsy. MRI can also identify complications of surgery or radiotherapy and monitor treatment response. Advanced imaging sequences provide valuable information about tumour and brain physiology, but clinical use is limited by extended scanning times and difficulties in processing and analysis. Brain eloquence classifications exist, but focus on adults with neurological deficits and are outdated. For the analysis of childhood tumours, limited numbers within tumour subgroups and the investigation of long-term outcomes necessitate using historical scans and/or multi-site collaboration. Variable imaging quality and differing acquisition parameters limit the use of segmentation algorithms and radiomic analysis. Harmonisation can standardise imaging in collaborative research, but can be challenging, while data-sharing produces further logistical challenges. Consequently, most research consists of small single-centre studies limited to regional analyses of tumour location. Technological advances reducing scanning times increase the feasibility of clinical acquisition of high-resolution standardised imaging including advanced physiological sequences. The RAPNO and SIOPE paediatric brain tumour imaging guidelines have improved image standardisation, which will benefit future collaborative imaging research. Modern machine learning techniques provide more nuanced approaches for integration and analysis of the complex and multifactorial data involved in cognitive outcome prediction.