Deep learning detection of informative features in tau PET for Alzheimer’s disease classification

BMC Bioinformatics

Table 2 All participants were partitioned into 5 subsets randomly, but every subset has the same ratio of CN and AD participants

	Train set		Test	Acc. r1	Epoch	Acc. r2	Epoch	Acc. r3	Epoch	Acc. r4	Epoch	Mean acc	SD
	Train	Val	Test	Acc. r1	Epoch	Acc. r2	Epoch	Acc. r3	Epoch	Acc. r4	Epoch	Mean acc	SD
fold1	78 (312)	28	26	85.7	23	92.9	24	89.3	21	85.7	27	88.4	3.0
fold2	78 (312)	28	26	96.2	20	96.2	21	92.3	21	88.5	22	93.3	3.2
fold3	80 (320)	26	26	100	35	92.3	28	88.5	27	88.5	30	92.3	4.7
fold4	80 (320)	26	26	92.3	24	88.5	31	88.5	38	88.5	29	89.4	1.7
fold5	80 (320)	26	26	92.3	50	80.8	36	96.2	35	92.3	34	90.4	5.8

One subset was selected for testing and the remaining four subsets were used for training. Among the four subsets for training, one subset (validation) was used without applying augmentation for tuning the weights of the layers without overfitting and the remaining three subsets were augmented. The numbers in parentheses are the training images after applying augmentation. The experiment was repeated four times for each fold (Acc. r1 ~ r4), and the mean accuracy was considered as the final accuracy of the fold. If the accuracy for the testing subset did not improve further within up to ten iterations, the training was stopped (epoch)

ISSN: 1471-2105