Meeting Banner
Abstract #2442

FreeSurfer Parcellation of Brains Containing Large Infarcts

Niharika Gajawelli1, Sinchai Tsao1, Darryl Hwang1, Bryce Wilkins1, Stephen Kriger2, Manbir Singh1

1Radiology and Biomedical Engineering, University of Southern California, Los Angeles, CA, United States; 2Center for Imaging of Neurodegenerative Diseases, VA Medical Center, San Francisco, CA, United States

FreeSurfer parcellation of the T1 weight human brain often fails in the presence of cerebral infarcts. This study presents a method of using voxels from the contralateral hemisphere to fill-in the cerebral infarct. The technique relied on SPM normalization including the normal and the hemispherically flipped versions of the brain, without any manual intervention other than an initial masking. A novel validation methodology was developed based on comparing ratios of FreeSurfer parcellated regions in a simulation study where the ground truth was known. Results show successful processing of 16 stroke subjects after filling, whereas processing in half failed before filling.

Keywords

accomplished accurate affected affine analyses applied automate boundaries brain brains caused cerebral circles classification clearly closeness closer combined comparing complete computed conduct containing corrected cortical deformation dementia develop deviation disease diseases disorders distal distortions distribution elderly engineering error even exactly example exhibits failed fails field finally fits five frequently ground hemisphere hemispheres highlight highly histograms homologous identification improvement indicate infarct infarcts inserted integrity intensive interaction interchanged intervention inverse investigated known lesions linear manual manually mapped mask medical neurological neuron normalization normalized often original outputs outside package patient patients play plot positioned presence previous problem processing progression quantitatively radiology realignment reconstruction recovering reduces reduction reflecting registering relatively remarkable requires respect respectively robust role scale similarity simulated simulation simulations smaller software space spatial step stroke structural structure structures studies subject subjected subjects successfully suggest suited surfer symmetrical template tissue truth typically unfilled validate validated validation vascular ventricles version vicinity volume volumes volumetric yellow