Fig 1. Proposed framework.

(A) HC subjects with normal CSF profile are identified from cutoff values calculated from CSF biomarkers distributions. (B) Longitudinal ROIs of these subjects are modelled using LME approach, variant (vr) and quasi-variant (qvr) ROIs and Y-intercepts (y₀) ROIs values are identified and then null models for both genders are built from these values by applying multivariate modelling. (C) qvr ROIs values of new HC, MCI and AD subjects are used within null models to infer the y₀ values of vr ROIs. Estimated ROIs values ($\widehat{y}$) at different ages are estimated by linear regression by using y₀ and β coefficients of age and education. Residuals are calculated as the difference $y-\widehat{y}$; and finally, SVM classifiers are trained for subject classification and addressing the early diagnosis problem: HC vs MCI, MCI vs. AD and HC vs AD. The full workflow of last two stages is applied separately for each gender.