Resting-state EEG already flags cognitive impairment
Resting-state EEG—brain activity recorded while a person sits quietly, without a specific task—has repeatedly been used to distinguish people with mild cognitive impairment (MCI) from healthy controls. In one study, a deep learning model classified subjective cognitive decline and MCI from eyes-closed resting EEG, reaching an AUC of 0.807 for separating the two groups [1]. AUC is a summary score where 0.5 is chance and 1.0 is perfect; 0.807 means the model ranked a randomly chosen impaired person above a randomly chosen less-impaired person about 81% of the time. The same study reported lower but still above-chance performance when separating healthy controls, subjective decline, and MCI together [1].
A separate machine-learning approach using resting-state EEG from electrodes over Broca's area—a language-related brain region—reported 90.91% accuracy in telling MCI from non-MCI individuals, based on phonemic verbal fluency assessment [5]. That figure comes from a single study and should be read as a benchmark within that dataset rather than a universal detection rate. In Parkinson's disease, a resting-state EEG index correlated strongly with the Montreal Cognitive Assessment (rho = 0.69) and detected cognitive impairment with 80.44% accuracy, 0.90 AUC, 79.52% sensitivity, and 81.14% specificity using only eight electrodes [2]. Sensitivity here means the share of impaired people correctly flagged; specificity means the share of unimpaired people correctly cleared.
What resting-state markers actually measure
Resting-state EEG does not test a person's performance on a mental challenge. Instead, it captures the brain's spontaneous activity patterns and the coordination between regions. In amnestic MCI, researchers found differences in resting-state networks compared with healthy elderly people—for example, increased delta and beta activity in the superior frontal gyrus and decreased alpha activity in a default-mode network region [4]. These are group-level differences in network usage, not a simple diagnostic test, and the study compared 125 aMCI patients with 70 healthy elderly participants [4].
That distinction matters for sensitivity. A resting-state marker can reveal network changes that a person cannot consciously compensate for, which may help early detection. But it may also miss subtle cognitive slippage that only appears when the brain is challenged. The evidence so far shows that resting-state signals carry useful diagnostic information, not that they capture every aspect of impairment.
Why the task-based comparison is still open
Task-based EEG records brain responses while a person performs a cognitive or sensory challenge, such as an attention or memory task. A review of prefrontal EEG markers in MCI discusses both resting-state and task-based analyses, noting that resting-state biomarkers have been consistently associated with cognitive decline [3]. However, that review does not report a direct sensitivity comparison showing task-based paradigms outperform resting-state ones [3].
The other studies here used resting-state designs exclusively: the deep learning classifier for subjective decline and MCI [1], the Parkinson's disease prediction index [2], the network analysis in amnestic MCI [4], and the Broca's-area machine-learning approach [5]. None of them tested a task-based paradigm head-to-head against resting-state markers in the same participants. Without such comparisons, the claim that task-based EEG is more sensitive remains plausible but unproven. Task-based designs may probe specific cognitive operations more directly, while resting-state designs are simpler to administer and have shown consistent group differences; which one wins on sensitivity likely depends on the impairment stage and the cognitive domain being tested.
Sources used in this answer
An attention-based deep learning approach for the classification of subjective cognitive decline and mild cognitive impairment using resting-state EEG
A deep learning model using resting-state EEG classified subjective cognitive decline versus mild cognitive impairment with an AUC of 0.807, and three-group classification with micro-AUC above 0.74.
Resting-state EEG Predicts Cognitive Impairment in Parkinson’s Disease (P6-11.015)
A resting-state EEG index using eight electrodes correlated with the Montreal Cognitive Assessment (rho = 0.69) and detected cognitive impairment in Parkinson's disease with 80.44% accuracy and 0.90 AUC.
Towards Early Alzheimer's Diagnosis: Prefrontal EEG Markers in Mild Cognitive Impairment from Resting State and Sensory-Attentional ERP Perspectives
A review discusses resting-state and task-based prefrontal EEG markers in mild cognitive impairment, noting resting-state biomarkers have been consistently associated with cognitive decline, but does not report a direct sensitivity comparison.
EEG Resting-State Functional Networks in Amnestic Mild Cognitive Impairment
Amnestic mild cognitive impairment patients showed altered resting-state EEG network activity compared with healthy elderly, including increased delta/beta in the superior frontal gyrus and decreased alpha in default-mode and attention network regions.
High-Performance Detection of Mild Cognitive Impairment Using Resting-State EEG Signals Located in Broca's Area: A Machine Learning Approach.
A machine learning approach using resting-state EEG from Broca's area channels achieved 90.91% accuracy in differentiating MCI from non-MCI individuals based on phonemic verbal fluency assessment.