Table 2.

Visual interpretation of VBM maps: Cohen kappa coefficients (and their 95%CI) for intra- and between-readers agreement and for the agreement between the consensus interpretation of the two readers with the ground truth diagnoses. The kappa values are given for the categorization according to three classes (AD versus FTLD versus normal) and for the detection of any neurodegenerative disease (AD or FTLD versus normal)

		Scanner-specific VBM	Multiple-scanner VBM	CNN-VBM
Intra-reader Reader 1	3 classes: AD vs FTLD vs normal	0.82 [0.73, 0.90]	0.79 [0.70, 0.89]	0.82 [0.73, 0.90]
	2 classes: AD or FTLD vs normal	0.84 [0.75, 0.94]	0.78 [0.67, 0.89]	0.79 [0.67, 0.90]
Intra-reader Reader 2	3 classes: AD vs FTLD vs normal	0.86 [0.78, 0.93]	0.94 [0.88, 1.00]	0.85 [0.78, 0.93]
	2 classes: AD or FTLD vs normal	0.88 [0.80, 0.96]	0.97 [0.93, 1.00]	0.84 [0.75, 0.93]
Between-reader	3 classes: AD vs FTLD vs normal	0.85 [0.77, 0.93]	0.84 [0.75, 0.93]	0.74 [0.64, 0.84]
	2 classes: AD or FTLD vs normal	0.88 [0.80, 0.96]	0.87 [0.79, 0.96]	0.71 [0.59, 0.83]
Reader consensus versus ground truth	3 classes: AD vs FTLD vs normal	0.77 [0.67, 0.87]	0.44 [0.32, 0.57]	0.72 [0.61, 0.82]
	2 classes: AD or FTLD vs normal	0.77 [0.65, 0.89]	0.37 [0.21, 0.52]	0.77 [0.65, 0.89]

AD Alzheimer’s disease, CNN convolutional neural network, CNN-VBM CNN-based VBM without reference to a normal database, FTLD frontotemporal lobar degeneration, multiple-scanner VBM conventional VBM with a mixed normal database comprising T1w-MRI images from multiple scanners as reference, scanner-specific VBM conventional VBM with a scanner- and sequence-specific normal database as reference, VBM voxel-based morphometry