Distinguishing between paediatric brain tumour types using multi-parametric magnetic resonance imaging and machine...

Area of Science:

• Radiology and Medical Imaging
• Machine Learning in Medicine
• Paediatric Oncology

Background:

• Accurate diagnosis of paediatric brain tumours is challenging, with magnetic resonance imaging (MRI) crucial for tumour-specific information.
• Diffusion-weighted imaging (DWI) and perfusion imaging are vital for non-invasive diagnosis but often rely on qualitative expert review.
• Existing quantitative studies are predominantly single-centre and single-modality, limiting generalizability.

Purpose of the Study:

• To integrate multi-centre diffusion and perfusion MRI data with machine learning (ML).
• To develop ML-based classifiers for discriminating between three common paediatric brain tumour types.
• To enhance the non-invasive diagnostic capabilities for paediatric brain tumours.

Main Methods:

• Combined multi-centre diffusion-weighted imaging (DWI) and perfusion imaging datasets.
• Utilized machine learning algorithms to build classifiers.
• Analyzed imaging features from both the tumour and the whole brain.

Main Results:

• Significant imaging features differentiating tumour types were identified in both tumour and whole brain DWI and perfusion data.
• Combining these diffusion and perfusion features yielded an optimal machine learning classifier.
• The developed classifier demonstrated over 80% predictive precision in tumour type discrimination.

Conclusions:

• Multi-centre diffusion and perfusion MRI data, when analyzed with machine learning, can effectively differentiate common paediatric brain tumours.
• This integrated approach offers a promising step towards more accurate and non-invasive diagnosis of paediatric brain tumours.
• The findings support the use of advanced clinical imaging techniques combined with ML for improved diagnostic outcomes.