Dallas Card1, Revital Nossin-Manor1,2,
John G. Sled3,4
1Diagnostic Imaging, the
Hospital for Sick Children, Toronto, Ontario, Canada; 2Neurosciences
& Mental Health, Research Institute, the Hospital for Sick Children,
Toronto, Ontaro, Canada; 3Physiology & Experimental Medicine,
Research Institute, the Hospital for Sick Children, Toronto, Ontario, Canada;
4Medical Biophysics, the University of Toronto, Toronto, Ontario,
Canada
Automated tissue classification in very preterm neonates presents unique challenges which commonly result in the misclassification of partial volume voxels. This abstract presents an alternative approach to this problem, based on the trimmed minimum covariance determinant method. This algorithm was used to segment the brains of 44 very preterm neonates into white matter, grey matter, cerebral spinal fluid, and two partial volume classes. Our results show clear differentiation between classes with the partial volume voxels correctly labeled in most cases. An inter-rater comparison involving 12 subjects confirmed its reliability, with all mean Dice coefficients greater than 0.8.