Alzheimer’s disease (AD) is the most common cause of dementia. Recommended clinical use of imaging has been limited to visual evaluations of brain MRI for ruling out “organic” causes of dementia such as stroke or tumor. No role of imaging currently exists for identifying mild cognitive impairment (MCI). Development of FDA cleared quantitative software allows for quantification of multiple brain regions.

To evaluate discriminative ability of a newer program for identifying AD from MCI and controls.

Volumetric 1.5 and 3.0T MRI brain scans (n = 1143 with mean age 74.6 years) were obtained from ADNI using standard protocols⁵. This cohort consisted of controls (n = 261), early mild cognitive impairment (EMCI, n = 310), late mild cognitive impairment (LMCI, n = 223), and AD (n = 349). Neuroreader  was used to compute brain volumes. Machine learning was done using cross-validated discriminant analysis algorithm in IBM SPSS Modeler (v. 18, Armonk, NY). Area under the curve (AUC) was generated for AD and MCI subgroups.

Regionally quantified volumetric MR imaging data separated AD from non-AD groups with AUC of 89%, 85% sensitivity, and 79% specificity. Automated volumetrics delineated LMCI from other groups with AUC of 72%, 70% sensitivity, and 62% specificity (Figure 1B). EMCI was distinguished form LMCI, AD, and controls groups with AUC of 80%, 76% sensitivity, and 70% specificity. Early MCI was distinguished from controls with 94% accuracy. Predictive regions delineating AD from MCI subgroups and controls included total CSF volume, hippocampal asymmetry and temporal lobe volumes.

Machine learning analysis of quantified brain regions on MR imaging provides good diagnostic delineation of AD from MCI subgroups and normal controls. Overlap between LMCI and AD and EMCI and controls may partially account for reduced diagnostic performance in MCI. Future studies will utilize longitudinal imaging for improved delineative outcomes.