Table 4.

5-fold cross-validation accuracy for the classification of ADNI participants, for all ModelE and all numbers of channels c. The number of channels corresponds to the number of 3D convolutional kernels used in each layer. ModelE28 refers to ModelE with 28 layers.

c	ModelE28	ModelE24	ModelE22	ModelE20	ModelE18	ModelE16	ModelE10	ModelE8
8	61.12%	70.09%	66.91%	58.12%	67.91%	64.53%	65.17%	49.40%
10	58.53%	73.90%	72.67%	65.31%	73.08%	62.80%	67.11%	56.41%
12	68.33%	60.37%	60.02%	78.08%	51.22%	72.51%	63.86%	53.41%
14	60.27%	71.09%	62.29%	62.74%	67.90%	59.41%	61.01%	59.56%
16	66.47%	72.61%	56.38%	65.47%	72.29%	75.87%	72.87%	54.35%
20	54.09%	61.21%	69.69%	62.79%	81.08%	75.09%	63.01%	54.38%
24	56.58%	69.46%	58.81%	64.16%	66.00%	71.09%	62.89%	57.56%
28	60.98%	68.86%	60.61%	66.30%	67.41%	66.60%	62.07%	56.01%