Domain-specific deficits in categorical perception unveil the neural architecture of auditory agnosia in dementia.
Auditory processing deficits are a prominent early feature of Alzheimer's disease (AD), yet whether they reflect generalized cognitive slowing or domain-specific neural degradation remains unclear. This study combined a categorical perception paradigm with high-temporal-resolution electroencephalography (EEG) to evaluate neural responses to distinct combinations of acoustic cues, rapid temporal dynamics (consonants) versus stable spectral configurations (lexical tones), across a continuum of healthy aging, mild cognitive impairment (MCI), and AD. Rather than a monolithic functional loss, behavioral analysis revealed a differential vulnerability: while psychomotor slowing was global, categorical precision collapsed specifically for rapid consonant cues but remained remarkably resilient for tones. Neurophysiologically, this divergence was underpinned by a distinct trajectory transitioning from successful compensatory hyper-activation in normal aging to a state of inefficient hyper-activation in clinical cohorts, characterized by sustained P300 amplitudes but severe temporal processing delays. Crucially, we demonstrated the clinical utility of these domain-specific electrophysiological signatures using a fivefold cross-validated machine learning approach. Advanced classification algorithms, notably the gradient boosting machine (GBM), distinguished participants with MCI from cognitively normal older adults with an area under the receiver operating characteristic curve (AUC) of 0.836. These findings reframe auditory deficits in dementia as a nuanced erosion of specific neural codes and highlight the efficacy of interpretable, EEG-based neurocomputational tools for the early clinical screening of pre-dementia states.