Table 1.

Comparison of a nonexhaustive selection of studies that perform prediction of future cognitive decline in the context of AD. Compared with the literature, which is often concerned with predicting discrete class assignment, our approach predicts rates of cognitive decline based on a wide selection of input features.

	Targets	Inputs	Analysis method
(Eskildsen et al., 2015)	MCI to AD conversion	Regional cortical thickness, nonlocal hippocampal morphological grading scores, clinical scores (MMSE, RAVLT), age	Linear discriminant analysis with multivariate feature selection
(Korolev et al., 2016)		Clinical scores (risk factors, clinical assessments, medication status), regional GM morphometry, (cortical and subcortical volumes, mean cortical thickness, standard deviation of cortical thickness, surface area, curvature), plasma proteomics biomarkers	Probabilistic multiple kernel learning (pMKL) with multivariate feature selection
(Gaser et al., 2013)		Estimated “brain age” score, baseline clinical scores, age, hippocampal volume	Cox regression, ROC analysis
Davatzikos et al., 2011)		Automated marker of atrophy (SPARE-AD), CSF biomarkers	SVM
(Bhagwat et al., 2018)	Membership to clusters of MMSE and ADAS-13 trajectories	Regional cortical thickness, APOE4 status, age and baseline clinical scores	Longitudinal siamese network
Current study	Rate of MMSE and CDR-SOB change	Mean cortical thickness, GM, WM and CSF total volumes, regional subcortical volumes, functional connectivity, age, APOE status, baseline clinical scores, demographic information, health status, neuropsychological assessment	Multitarget Random Forest