Radiomics in paediatric neuro-oncology: A multicentre study on MRI texture analysis

Area of Science:

• Medical Imaging
• Oncology
• Artificial Intelligence

Background:

• Paediatric brain tumours are the leading cause of cancer deaths in UK children.
• Current non-invasive diagnostic methods for paediatric brain tumours lack routine clinical integration.
• Radiomics offers potential for quantitative image feature analysis in tumour characterization.

Purpose of the Study:

• To evaluate the application of MRI texture analysis (TA) for classifying paediatric brain tumours.
• To assess the cross-centre transferability of TA models using support vector machines (SVM).
• To investigate feature selection for generalizable classifiers and address class imbalance.

Main Methods:

• A multicentre study utilized conventional MRI scans from paediatric patients.
• A supervised classification framework trained a support vector machine (SVM) with 3D textural attributes.
• Rigorous pairwise testing assessed SVM performance on unseen datasets for transferability.

Main Results:

• Optimal area under the receiver operating characteristic curve ranged from 76% to 86%, indicating transferable tumour information.
• Feature selection revealed consistent texture aspects enhanced across different hospital imaging data.
• Addressing class imbalance improved the SVM's data characterization capabilities.

Conclusions:

• 3D TA on conventional MRI is a viable tool to support the diagnostic classification of paediatric brain tumours.
• The study demonstrates the potential for non-invasive, accurate tumour characterization in paediatric oncology.
• Texture analysis shows promise for improving diagnostic accuracy and clinical decision-making in paediatric neuro-oncology.