Neurotransmitter-based machine-learning model for distinguishing Alzheimer's disease and mild cognitive impairment.

ObjectiveAlzheimer's disease and mild cognitive impairment involve brain atrophy, but neurotransmitter changes and their clinical implications are not well defined. This study aimed to examine the relationship between gray matter atrophy and neurotransmitter distributions and to build machine-learning models using gray matter-neurotransmitter co-localization as features.MethodsAmong 262 participants from the Alzheimer's Disease Neuroimaging Initiative (140 with Alzheimer's disease, 50 with mild cognitive impairment, and 72 controls), we used structural magnetic resonance imaging and voxel-based morphometry (family-wise error < 0.05), and JuSpace toolbox was used to assess the spatial correlation between gray matter atrophy and 13 neurotransmitter maps. We applied a train/validation/fixed test split (the test set was never used for selection or training); features were screened by univariate regression and least absolute shrinkage and selection operator regression, and models trained with nested 10-fold cross-validation were evaluated by the area under the receiver operating characteristic curve.ResultsBoth Alzheimer's disease and mild cognitive impairment showed gray matter loss in temporal, frontal, and cingulate areas. Atrophy was correlated with serotonergic, dopaminergic, and glutamatergic systems (false-discovery rate < 0.05). In mild cognitive impairment, reduced metabotropic glutamate receptor 5/μ-opioid receptor-gray matter correlation was associated with higher depression scores (r = -0.44, p = 0.001; r = -0.44, p = 0.001). The Random Forest model achieved an area under the receiver operating characteristic curve of 0.821, and Shapley additive explanations analysis confirmed key feature contributions.ConclusionNeurotransmitter-linked gray matter changes contribute to the pathology of Alzheimer's disease and mild cognitive impairment. The machine-learning model accurately differentiates these conditions, suggesting its utility for early diagnosis and disease staging.