Fast and robust extraction of hippocampus from MR images for diagnostics of Alzheimer's disease

Abstract

Assessment of temporal lobe atrophy from magnetic resonance images is a part of clinical guidelines for the diagnosis of prodromal Alzheimer's disease. As hippocampus is known to be among the first areas affected by the disease, fast and robust definition of hippocampus volume would be of great importance in the clinical decision making. We propose a method for computing automatically the volume of hippocampus using a modified multi-atlas segmentation framework, including an improved initialization of the framework and the correction of partial volume effect. The method produced a high similarity index, 0.87, and correlation coefficient, 0.94, with semi-automatically generated segmentations. When comparing hippocampus volumes extracted from 1.5T and 3T images, the absolute value of the difference was low: 3.2% of the volume. The correct classification rate for Alzheimer's disease and cognitively normal cases was about 80% while the accuracy 65% was obtained for classifying stable and progressive mild cognitive impairment cases. The method was evaluated in three cohorts consisting altogether about 1000 cases, the main emphasis being in the analysis of the ADNI cohort. The computation time of the method is about 2 minutes on a standard laptop computer. The results show a clear potential for applying the method in clinical practice.

Fig. 1

The segmentation pipeline showing also transformations between the unseen data, template and atlas spaces.

Fig. 2

Bland–Altman plot for semi-automatically and automatically defined volumes in the ADNI N = 340 cohort. Horizontal lines show the mean ± 2*standard deviation.

Fig. 3

Segmentation results of six cases from which the first two show one cognitively normal (a) and one AD case (b) with average segmentation accuracy (SI≈0.87) and the rest four (c–f) the cases with the lowest SI values from the cohort N = 340. The top and bottom rows show the semi-automatic and automatic, respectively, segmentations superimposed on the image. On the locations where the thickness of the yellow line is higher than a voxel, the surface and the image plane are partially parallel and the surface cross-sects several neighboring voxels. The left and right hippocampi are shown in a sagittal view and in a transaxial view for each case. The similarity index is reported both for the left and right sides, the volume of hippocampi when using semi-automatic and automatic segmentations (S/A), and the ADNI classification of the patient (C): a) SI(L/R) = 0.844/0.868, V(S/A) = 4.3/4.4 ml, C = CN, b) SI(L/R) = 0.852/0.892, V(S/A) = 3.1/3.5 ml, C = AD, c) SI(L/R) = 0.743/0.671, V(S/A) = 2.8/2.9 ml, C = PMCI, d) SI(L/R) = 0.702/0.817, V(S/A) = 4.5/4.5 ml, C = PMCI, e) SI(L/R) = 0.654/0.697, V(S/A) = 4.1/3.8 ml, C = not known, and f) SI(L/R) = 0.635/0.863, V(S/A) = 3.3/3.3 ml, C = not known.

Fig. 4

Bland–Altman plot for hippocampus volumes computed using 1.5 T and 3 T images in the ADNI N = 181 cohort. Horizontal lines show the mean ± 2*standard deviation.

Fig. 5

Boxplots computed for volumes of hippocampus (CN = cognitively normal, SMCI = stable MCI-patient, PMCI-progressive MCI-patient and AD = Alzheimer's disease patient) using semi-automatic (left) and automatic (center) segmentations in the ADNI cohort N = 321 and automatic segmentations in the ADNI cohort N = 776 (right).

Fig. 6

ROC-curve for the classification performance using semi-automatically and automatically generated volumes of hippocampus in the cohort N = 321.

Fig. 7

Boxplots computed for volumes of hippocampus (CN = cognitively normal, SMCI = stable MCI-patient, PMCI-progressive MCI-patient and AD = Alzheimer's disease patient) using the Kuopio (N = 106) and GEHC (N = 72) cohorts.

Fig. 8

Probabilistic atlases used as spatial priors in the expectation maximization segmentation: a) original MR image, and the probabilistic atlas of b) CSF, c) gray-matter, d) white-matter and e) hippocampus.