RSNA 2014 

Abstract Archives of the RSNA, 2014


SSK16-08

Alzheimer’s Disease Diagnostic Performance of a Multi-Atlas Hippocampal Segmentation Method Using the Harmonized Hippocampal Protocol 

Scientific Papers

Presented on December 3, 2014
Presented as part of SSK16: Neuroradiology (Advanced Neuroimaging of Alzheimer's Disease)

Participants

Cecilie Benedicte Anker MSc, Abstract Co-Author: Nothing to Disclose
Lauge Sorensen, Abstract Co-Author: Research funded, Biomediq A/S
Akshay Pai, Abstract Co-Author: Nothing to Disclose
Mark Lyksborg PhD, MSc, Abstract Co-Author: Nothing to Disclose
Martin Lillholm PhD, Abstract Co-Author: Employee, Biomediq A/S Shareholder, Biomediq A/S
Mads Nielsen PhD, Presenter: Stockholder, Biomediq A/S Research Grant, Nordic Bioscience A/S Research Grant, SYNARC Inc Research Grant, AstraZeneca PLC
Knut Conradsen, Abstract Co-Author: Nothing to Disclose
Rasmus Larsen, Abstract Co-Author: Nothing to Disclose

PURPOSE

Hippocampal volumetry is the most widely used structural MRI biomarker of Alzheimer’s disease (AD), and state-of-the-art, automatic hippocampal segmentation can be obtained using longitudinal FreeSurfer. In this study, we compare the diagnostic AD performance of a single time point, multi-atlas method using the Harmonized Hippocampal Protocol (HHP) to FreeSurfer (FS).

METHOD AND MATERIALS

Baseline and month 12 MRI scans from the “complete annual year 2 visits” 1.5-T standardized ADNI dataset were used [169 normal controls (NC), 234 mild cognitive impaired (MCI), 101 AD]. A multi-atlas, affine registration, patch-based segmentation method (MRP) using 40 HHP segmentations in the atlas (12 NC, 11 MCI, 17 AD) was applied to segment the hippocampi. Static- and longitudinal FS (v5.1.0, default parameters) were also applied to segment the hippocampi. Atrophy rate calculated as percent volume change from baseline to month 12 was estimated for the three methods, and diagnostic performance was evaluated using the area under the receiver operating characteristic curve (AUC) of pairwise diagnostic group comparisons.

RESULTS

Mean (SD) atrophy rates were as follows (MRP / static FS / longitudinal FS): NC -0.86 (2.46) / -1.39 (5.41) / -1.63 (2.54), MCI -2.38 (3.28) / -3.69 (5.48) / -3.25 (3.53), AD -4.23 (3.07) / -4.29 (5.32) / -4.83 (3.74). Diagnostic performances were as follows (AUC; MRP / static FS / longitudinal FS): NC vs. MCI 0.65 / 0.67 / 0.64, NC vs. AD 0.80 / 0.69 / 0.76, MCI vs. AD 0.66 / 0.53 / 0.62. The MRP AUC was significantly larger (DeLong) than the static FS AUC for NC vs. AD and MCI vs. AD. In the remaining pairwise group comparisons, MRP AUCs did not differ significantly from FS AUCs.    

CONCLUSION

The MRP method discriminated AD from either NC or MCI significantly better than static FS, and it was as good as longitudinal FS, which exploits information from both time points simultaneously. Moreover, the standard deviation of the atrophy rate was comparable to that of longitudinal FS, emphasizing longitudinal robustness of segmentations of the proposed method. The combination of MRP and HHP is a robust and fast alternative to FreeSurfer, especially in a setting with many time points. 

CLINICAL RELEVANCE/APPLICATION

Unlike longitudinal FS, the MRP method calculates final atrophy estimates after each visit. Adding the comparable performance, the proposed method is a robust alternative for clinical trials. 

Cite This Abstract

Anker, C, Sorensen, L, Pai, A, Lyksborg, M, Lillholm, M, Nielsen, M, Conradsen, K, Larsen, R, Alzheimer’s Disease Diagnostic Performance of a Multi-Atlas Hippocampal Segmentation Method Using the Harmonized Hippocampal Protocol .  Radiological Society of North America 2014 Scientific Assembly and Annual Meeting, - ,Chicago IL. http://archive.rsna.org/2014/14012721.html