Premium
P3‐382: METHODS TO IMPROVE SPM12 TISSUE SEGMENTATIONS OF OLDER ADULT BRAINS
Author(s) -
Schwarz Christopher G.,
Gunter Jeffrey L.,
Ward Chadwick P.,
Kantarci Kejal,
Senjem Matthew L.,
Petersen Ronald C.,
Knopman David S.,
Jack Clifford R.
Publication year - 2018
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2018.06.1744
Subject(s) - voxel , normalization (sociology) , segmentation , spatial normalization , atrophy , medicine , pattern recognition (psychology) , white matter , artificial intelligence , anatomy , computer science , pathology , magnetic resonance imaging , radiology , sociology , anthropology
PD, and FLAIR MRI were acquired in two separate sessions from fourteen subjects. The published 4-sequence algorithm (mMAPS) was applied and three visual raters scored single slice sections. T1 and T2 were used for single sequence segmentation (MAPS). Inside a white matter (WM) mask, voxels were a) inverse frequency ranked (FRANK4) and b) mean difference scores (DIFF5) calculated (Figure 1). For T1, a voxel (with intensity i) was considered a possible ePVS if [FRANK4>0.95, DIFF5>.15*i]. For T2, [FRANK4<0.95, DIFF5<.15*i]. Each 5mm cluster was further subjected to morphological filtering on linearity and width; linearity was calculated as percent explained variance of the primary eigenvector of each cluster, width was calculated as the intersection of planes normal to that vector and the edges of each cluster.Results: There was a high correspondence between visual ePVS counts and mMAPS, T1, and T2 counts over all subjects; all methods were significantly correlated over repeated measurements in the single visually rated slice and over the whole brain (Figure 2). SNR and CNR of segmented ePVS using single sequence methods and all statistical results are available in Table 1. Conclusions: While consultation of several sequence weightings is preferred to ensure that a putative ePVS is isointense to CSF on all sequences, the single sequence automated method of segmentation, based largely on refined morphological constraints, is highly correlated to expert visual counts and previously published automated methods. This method may be applicable to large single sequence datasets such as ADNI, and would add valuable information regarding location and morphology of ePVS.