TABLE I.

ADNI Dataset. Top: Comparison across methods on NC vs. AD classification; Bottom: Balanced Classification Accuracy of the proposed method dependent on the number of visits

Method	BACC ± std	SEN	SPE	AUC
CNN	86.1 ± 1.83	86.8	85.4	91.3
CNN+AP	87.8 ± 2.55	88.2	87.4	91.3
CNN+RNN	89.5 ± 1.41	88.7	90.3	91.4
CNN+RNN+LP	90.0 ± 1.44	89.2	90.8	91.5
CNN+RNN+LP+CL	90.4 ±1.51	88.9	91.9	91.5
CNN+biRNN [7]	89.6 ± 1.48	86.8	92.4	91.5
CNN+tRNN [27]	89.7 ± 1.67	88.8	90.6	91.4

	Visits	NC/AD	BACC ± std
CNN+RNN+LP+CL	1+ visit	188/169	89.8 ± 1.48
	2+ visits	163/103	90.1 ± 1.53
	3+ visits	138/78	90.6 ± 1.59
	4+ visits	109/52	90.6 ± 1.46
	5+ visits	73/0	91.8 ± 1.49