Meeting Banner
Abstract #4794

Automated lesion detection from multimodal brain MRI using Markov random fields and random forest

Jhimli Mitra 1 , Soumya Ghose 1 , Pierrick Bourgeat 1 , Olivier Salvado 1 , Stephen Rose 1 , Bruce Campbell 2 , Alan Connelly 3 , Susan Palmer 4 , Leeanne Carey 4 , and Jurgen Fripp 1

1 The Australian e-Health Research Centre, CSIRO Computational Informatics, CSIRO Preventative-Health Flagship, Herston, QLD, Australia, 2 Department of Medicine and Neurology, Royal Melbourne Hospital, Parkville, VIC, Australia, 3 Department of Radiology, Royal Melbourne Hospital, Parkville, VIC, Australia, 4 The Florey Institute of Neuroscience and Mental Health, Parkville, VIC, Australia

We present an automated method to delineate chronic ischemic stroke lesions, white-matter hyperintensities and other secondary lesions from multimodal MRI. Accurate delineation of such lesions is crucial in analyzing the structure-function relationships of the brain post-stroke and critical in the management of stroke patients. The method firstly employed a maximum aposteriori-Markov field based segmentation of the probable lesion areas from hyperintense regions of FLAIR images. Then features of these lesion areas from the multimodal MRI were used to train/test a random forest classifier. The performance was evaluated on 36 stroke patients (mean Dice 0.60+/-0.12, volume correlation 0.76)

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

Join Here