Meeting Banner
Abstract #1062

Paediatric brain lesion classification using 3T MRS: Comparison of different pattern recognition techniques, A multi-centre study.

Niloufar Zarinabad1,2, Laurence J J Abernethy3, Dorothee P Auer4,5,6, Theodoros N Arvanitis2,7, Simon Bailey8, Nigel P Davies1,2,9, Daniel Rodriguez Gutierrez 4,5, Richard G. Grundy 4, Tim Jaspan 4,6, Dipayan Mitra 10, Paul S Morgan4,6,11, Barry Pizer 12, Martin Wilson1, Lesley MacPherson2, and Andrew Peet1,2

1University of Birmingham, Birmingham, United Kingdom, 2Birmingham children hospital, Birmingham, United Kingdom, 3Department of Radiology, Alder Hey Children's NHS Foundation Trust, 4The Children‘s Brain Tumour Research Centre, University of Nottingham, Nottingham, United Kingdom, 5Radiological Sciences, Department of Clinical Neuroscience, University of Nottingham, Nottingham, United Kingdom, 6Neuroradiology, Nottingham University Hospital, Queen’s Medical Centre, Nottingham, United Kingdom, 7Institute of Digital Healthcare, WMG, University of Warwick, Coventry, United Kingdom, 8Paediatric Oncology Department, Great North Children’s Hospital, Newcastle upon Tyne, United Kingdom, 9Department of Imaging and Medical Physics, University Hospitals Birmingham NHS Foundation Trust, Birmingham, United Kingdom, 10Neuroradiology Department, Newcastle upon Tyne Hospitals, Newcastle upon Tyne, United Kingdom, 11Medical Physics, Nottingham University Hospital, Queen’s Medical Centre, Nottingham, United Kingdom, 12Department of Paediatric Oncology, Alder Hey Children's NHS Foundation Trust, Liverpool, United Kingdom

The purpose of the study was to investigate the discriminative potential of metabolites obtained from 3T scanners in classifying paediatric posterior fossa brain tumours by comparing performance of three different pattern recognition techniques on a multicentre data set. A total of 52 paediatric patients with cerebellar tumours (16 Medulloblastomas, 31 Pilocytic Astrocytomas and 5 Ependymomas) were scanned using PRESS, TE 30-46 ms, across 4 different hospitals. Achieved balanced classification accuracy were 88% with random-forest, 84 % for the support-vector-machine and 81% for naïve-bays classifier. The achieved accuracy was better than the balanced accuracy previously reported for multi-centre datasets at 1.5T.

This abstract and the presentation materials are available to members only; a login is required.

Join Here