Early diagnosis of Alzheimer’s disease using machine learning: a multi-diagnostic, generalizable approach

Table 1 Summary of diagnostic classification experiments

Experiment	Training set	Testing set	GT metrics
Using ADNI: testing generalizability across acquisition protocols
A1	ADNI MPRAGE	ADNI MPRAGE	No
A2	ADNI IR-SPGR	ADNI IR-SPGR	No
A3	ADNI MPRAGE	ADNI IR-SPGR	No
A4	ADNI IR-SPGR	ADNI MPRAGE	No
A5	ADNI IR-SPGR and ADNI MPRAGE	ADNI IR-SPGR and ADNI MPRAGE	No
A6	ADNI MPRAGE	ADNI MPRAGE	Yes
A7	ADNI IR-SPGR	ADNI IR-SPGR	Yes
A8	ADNI MPRAGE	ADNI IR-SPGR	Yes
A9	ADNI IR-SPGR	ADNI MPRAGE	Yes
A10	ADNI IR-SPGR and ADNI MPRAGE	ADNI IR-SPGR and ADNI MPRAGE	Yes
Using ADNI and OASIS: testing generalizability across datasets
B1	ADNI MPRAGE	ADNI MPRAGE	No
B2	OASIS (MPRAGE)	OASIS (MPRAGE)	No
B3	OASIS (MPRAGE)	ADNI MPRAGE	No
B4	ADNI MPRAGE	OASIS (MPRAGE)	No
B5	ADNI MPRAGE and OASIS (MPRAGE)	ADNI MPRAGE and OASIS (MPRAGE)	No
B6	ADNI MPRAGE	ADNI MPRAGE	Yes
B7	OASIS (MPRAGE)	OASIS (MPRAGE)	Yes
B8	OASIS (MPRAGE)	ADNI MPRAGE	Yes
B9	ADNI MPRAGE	OASIS (MPRAGE)	Yes
B10	ADNI MPRAGE and OASIS (MPRAGE)	ADNI MPRAGE and OASIS (MPRAGE)	Yes

‘GT metrics’ refers to whether graph theory metrics were given as an input to classifiers, regardless of whether the classifier selected them or not. For each of the experiments A1 through A10, 4 classifiers were built (“HC vs. MCI”; “HC vs. AD”; “MCI vs. AD”; and “HC vs. MCI vs. AD”). For each of the experiments B1 through B10, only 1 classifier was built (“HC vs. AD”, as OASIS did not have MCI subjects available)

ISSN: 1758-9193