Introduction

"Radiomics" and "Radiogenomics"

The high amount of information contained within medical images has led their inclusion in diagnostic protocols for many neurological diseases. The medical images are interpreted by a radiologist, but computer aided detection (CAD) can also be used to reduce observational oversights \cite{Castellino2005} and increase diagnostic accuracy, particular in cancer. However, there remains a large amount of complementary information within the medical images which could potentially aid with diagnosis. "Radiomics" attempts to capture and utilise this information to further aid diagnosis. Radiomics is the process of extracting a large number of image features from numerous independent cases, and using machine learning approaches to correlate these with known clinical and biological endpoints. The acquired model can then be used to predict unknown endpoints in further cases. 
Further, it is known that the assessment of gene expression profiles in cancerous tumours can improve classification rates, and therefore allow the patient to receive the most adequate treatment \cite{Collins2004}. However, acquiring the gene expression profiles requires invasive medical procedures, where samples are often collected during surgical resection. As radiographic imaging is part of clinical care and are readily available, it provides a potential surrogate for gene expression information. This approach describes the field of "radiogenomics", and has been used successfully in a number of oncological studies \cite{Qian2016,Hu2017,Kickingereder2016,Liu2017}.

Medulloblastoma

Medulloblastoma (MB) is a form of malignant tumour of the central nervous system (CNS) which are seen predominantly in children, although they can also occur in adults. They are the most common form of malignant brain CNS tumour and posterior fossa tumour in childhood, accounting for 12-15% and 38% of diagnosed cases respectively. The incidence rate of MB in children under the age of 9 is 10.52 per million in the U.S., where diagnosis in this rage accounts for 78% of all cases. The vast majority on MBs are located within the cerebellum and originate from the midline cerebellar vermis. Treatment methods include surgical resection, adjuvant chemotherapy and radiotherapy, and craniospinal irradiation, which have been shown to increase survival rate by 70-80%, despite the association with severe morbidity.
The heterogeneity of MB has led to their classification into subgroups to most effectively treat each tumour type (Table \ref{490314}). Modern clinical classification of MB assess the risk of the tumour and categorise them as "standard risk" or "high risk" in a criteria defined by the North American Children's Oncology Group and International Society of Paediatric Oncology, which takes into account patient age, metastasis and remaining tumour following surgical resection. MB are also classified by histological means to a criteria defined by the World Health Organisation (WHO) classification system (2016) into four subgroups: classic MB, desmoplastic, MB with extensive nodularity (MBEN), and large cell or anaplastic.
In 2002, Pomeroy et al., (2012)  \cite{Pomeroy2002} demonstrated the CNS tumours (in particular MB) can be differentiated using their gene expression profiles at diagnosis. Group have since conducted similar studies and produced similar results (references). This has led to the WHO updating their MB classification system (reference) to include the 4 identified subgroups; Wingless (WNT), Sonic Hedgehog (SHH), Group 3 and Group 4. WNT and SHH are named due to their signalling pathways, however less is understood surrounding Group 3 and Group 4 so generic names are given (reference).