Suchada Tantisatirapong, Nigel P. Davies, Lawrence Abernethy, Dorothee P. Auer, Chris A. Clark, Richard Grundy, Tim Jaspan, Darren Hargrave, Lesley MacPherson, Martin O. Leach, Geoff S. Payne, Barry L. Pizer, Andrew C. Peet, and Theodoros N. Arvanitis
Brain tumours, semi-automated segmentation, diffusion and conventional MRI, texture analysis
Primary brain tumours are the most common solid tumours found in children and are an important cause of morbidity and mortality. Magnetic resonance imaging (MRI) is commonly used for non-invasive early-detection, diagnosis, delineation of tumours for treatment planning and assessment of post treatment changes. Different MRI modalities provide complementary contrast of tumour tissues, which can have varying degrees of heterogeneity and diffusivity in different tumour types. A variety of texture analysis methods have been shown to reveal tumour histological types. It is hypothesized that textural features, based on conventional and diffusion MRI modalities, would differentiate the characteristics of tumours. Tumour extraction is also a significant procedure needed to obtain a true tumour region. Semi-automated segmentation methods were applied, in comparison with the gold standard of manual segmentation by an expert, in order to speed up a manual segmentation approach and reduce any bias effects. In this study, we present an automatic processing pipeline for the characterization of brain tumours, based on texture analysis. We apply this to a multi-centre dataset of paediatric brain tumours and investigate the accuracy of tumour classification, based on textural features of diffusion and conventional MR images.
Important Links:
Go Back
