[ic‐p‐103]: Application of a Multi‐atlas Segmentation Tool to Hippocampus, Ventricle, and Whole Brain Segmentation
Year: 2017
Type: article
Abstract: MRI-derived hippocampus and ventricle volumes are widely used as key biomarkers as well as to aid the quantitative analysis of amyloid/tau PET scans for normal aging and Alzheimer's Disease (AD). Here we present a use-case of a multi-atlas segmentation (MAS) tool implemented in inviCRO's VivoQuant™ software platform for fully automated hippocampus, ventricles and whole brain segmentation and evaluate its performance against manual tracing and FreeSurfer. Preprocessing includes denoising, bias field correction, AC-PC alignment and intensity normalization. 66 labeled MRIs (reference library) were affinely-registered to target images and a bounding box was generated around the ROI. Next, regions within the bounding box of the reference MRIs were warped to the target image using deformable registration. The warped labels corresponding to the 14 reference MRIs best match to target were fused. The performance was evaluated using 2 ADNI datasets. Dataset1 comprised of 132 MRIs with hippocampus manually traced based on the EADC-ADNI HarP protocol. Dataset2 comprised of 500 MRIs collected longitudinally from 126 subjects. The atrophy rate was estimated as the slope of a linear fit to all the volumes estimated for each subject and converted to percent change per year from baseline volume. The VivoQuant™-MAS derived atrophy rate were compared against Freesurfer output. For dataset1, the Dice index between VivoQuant™-MAS derived segmentation and manual tracings were 0.81±0.05 and 0.82±0.05 for left and right hippocampus. For dataset2, normal aging and disease progression-related reduction in hippocampus and whole-brain volume in tandem with increase in ventricle volume is observed with both methods. Using the VivoQuant™-MAS method (Table 1), the AD group exhibited a greater atrophy of a) left hippocampus relative to MCI, LMCI and Normal groups and b) right hippocampus relative to MCI and Normal groups. In contrast with Freesurfer (Table 2) only the comparison of left hippocampus atrophy in AD and Normal groups reached statistical significance. VivoQuant™-MAS, a fully automated segmentation tool, achieved high overlap with the EADC-ADNI validation dataset and performed favorably relative to Freesurfer in sensitively quantifying regional brain volume changes. Efforts are underway to deploy VivoQuant™-MAS in a CFR21-Part11 compliant manner to support registration trials.
Source: Alzheimer s & Dementia
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4231129543