Template impact in nucleus basalis segmentation

AbstractBackgroundThe Nucleus Basalis of Meynert (NB), located within the basal forebrain sublenticular region, is a cholinergic center with neurons providing cholinergic afferents to the entire neocortex. As posterior NB provides cholinergic innervation to the temporal pole and superior temporal cortex, this correlates well with memory loss and language impairment in AD.NB volume can serve as a potential emerging biomarker using volumetric MRI. One method consists in using DARTEL normalization to register images to a basal forebrain atlas. Our aim is to study the influence of template content and determine what is sufficient in practice for accurate evaluation, in the context of an on‐going multi‐center trial where the final dataset would not be available until study end.Methods96 good quality 3D T1‐weighted datasets were randomly chosen among screening scans of an on‐going AD trial (TRx‐237‐039/NCT03446001), with equal distribution of datasets from GE, Philips and Siemens, 1.5T/3T scanners (16 each). Separate DARTEL templates were created from the full list of 96 subjects, from subsets of 24, 48 and 72 subjects, and from independent sets of 20 and 96 random ADNI AD subjects to assess comparability.Processing followed a standard SPM12 DARTEL approach. 3DT1 images were initially segmented to create gray/white matter tissue maps and subsequently registered to create a template image. Derived flow fields were applied to warp and smooth all subjects to MNI space while preserving regional signal. Final subject maps were resliced to match a basal forebrain ROI.Volumes were compared with two‐sided paired t‐tests. Correlations across volumes were assessed with Pearson’s coefficient. Mean absolute percent difference to full 96‐template was calculated.ResultsVolumes obtained from either template were all highly and significantly correlated to each other (r>0.99, p<0.001). T‐tests showed significant differences (p<0.001), respectively resulting in mean absolute differences to the full template of 2.3% for the 24‐template, 0.8% for the ADNI‐20‐template, 1.6% for the ADNI‐96‐template and 0.3% for all others.ConclusionTemplate choice does impact measurements significantly. Choosing a sufficiently large sample is important when building the template, to avoid biasing overall results. A common template is essential for proper comparison of separate cohorts with this method.