Multi-parametric classification of traumatic brain injury patients using automatic analysis of quantitative MRI scans

  • Authors:

  • Benjamin S. Aribisala;Christopher J. A. Cowie;Jiabao He;Joshua Wood;A. David Mendelow;Patrick Mitchell;Andrew M. Blamire

  • Affiliations:

  • Institute of Cellular Medicine, Newcastle MR Centre, Newcastle University, UK;Institute of Cellular Medicine, Newcastle MR Centre, Newcastle University, UK;Institute of Cellular Medicine, Newcastle MR Centre, Newcastle University, UK;Institute of Cellular Medicine, Newcastle MR Centre, Newcastle University, UK;Dept of Neurosurgery, Newcastle General Hospital, Newcastle upon Tyne, UK;Dept of Neurosurgery, Newcastle General Hospital, Newcastle upon Tyne, UK;Institute of Cellular Medicine, Newcastle MR Centre, Newcastle University, UK

  • Venue:
  • MIAR'10 Proceedings of the 5th international conference on Medical imaging and augmented reality
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Traumatic brain injury (TBI) is ranked as the fourth highest cause of death in the developed world. The majority of patients sustain mild TBI, and a significant number suffer persistent neuropsychological problems. Conventional neuroimaging methods (CT, MRI) do not reveal abnormalities consistent with the cognitive symptoms. Imaging methods offering prognostic information in acutely injured patients are therefore required. Here we applied advanced quantitative MRI techniques (T1, T2 mapping and diffusion tensor MRI) in 24 mild TBI patients and 20 matched controls. We applied a support vector machine (SVM) to classify the quantitative MRI data. Univariate classification was ineffective due to overlap between patient and control values, however multi-parametric classification achieved sensitivity of 88% and specificity of 75%. Future work incorporating neuropsychological outcome into SVM training is expected to improve performance. These results indicate that SVM analysis of multi-parametric MRI data is a promising approach for predicting prognosis following mild TBI.