Perceptual Awareness in Alzheimer Disease Measured by EEG and fMRI in Huntley and Mediano's 2026 Study

Jonathan Huntley of the University of Exeter and University College London led a medRxiv preprint posted August 31, 2026, “A multimodal investigation of perceptual awareness in Alzheimer’s disease” (doi:10.64898/2026.08.27.26356661), with Benjy Barnett, Daniel Bor, Marco Mancuso, Pedro A. M. Mediano, Lorina Naci, Stephen Fleming, Giacomo Bertazzoli, Linda Clare, Adrian M. Owen, Lorenzo Rocchi, and Robert Howard. The study asks a question every consciousness theory must survive: what happens to the physiological signatures of perceptual awareness as the brain fails and the patient loses the ability to report.

The direct answer is that the signatures fade with the disease. Healthy older controls showed the expected cortical responses in both EEG and fMRI. Patients with mild to moderate Alzheimer disease showed significantly reduced visual awareness negativity, late positive responses, and fronto-parietal activation. Patients with severe disease, who were behaviourally minimally responsive, showed limited evidence of fronto-parietal awareness markers, though the authors caution this may partly reflect attentional and task insensitivity rather than absent awareness.

The paradigm and the markers

The team used a visual masking paradigm, the standard way to separate seeing from not seeing while holding the stimulus constant. Electroencephalography captured two event-related components. The visual awareness negativity (VAN) is a posterior negative deflection that tracks whether a stimulus entered awareness. The late positive (LP) component follows it and is associated with conscious access and report. Functional MRI measured activation in visual cortex and in fronto-parietal networks, the circuit that global workspace and recurrent processing accounts both treat as central to conscious access. A complex audio-visual movie task probed the same fronto-parietal networks under naturalistic viewing.

Why non-verbal measurement is the hard case

Every consciousness indicator is validated against self-report in healthy adults, yet the populations where measurement is most needed, advanced dementia, disorders of consciousness, infants, and non-verbal systems generally, are exactly the populations that cannot report. Huntley’s team meets that problem with physiology rather than language. The design parallels the logic of the perturbational complexity index, which reads consciousness level from the complexity of the brain’s response to stimulation rather than from behaviour, and of intracranial and scalp recordings that arbitrate between competing indicators in unresponsive patients.

The study is also a bridge between two theory families. The fronto-parietal activation measures speak to global workspace theory, where conscious access is broadcasting across a frontal-parietal architecture. The complexity and masking measures speak to recurrent processing and integrated-information style accounts, where awareness lives in the dynamics of posterior cortex. Reading both in the same patients, in the same sessions, is what makes the preprint more than a dementia paper. Pedro Mediano’s own information-decomposition work, PhiID, supplies the formal vocabulary for what such multimodal complexity measures are actually decomposing, and Stephen Fleming’s metacognition research frames the confidence side of the same architecture. The prefrontal-parietal circuit itself was set out for this site in Daniel Bor’s earlier synthesis with Anil Seth, and the study measures that circuit as it fails.

Comparison to The Consciousness AI

The Consciousness AI project treats consciousness as an emergent property of organization and builds measurement infrastructure on that footing. Huntley’s result constrains any such program in a specific way. The EEG and fMRI correlates of awareness are not static properties of a brain region. They degrade as the tissue’s dynamics degrade, in step with disease progression. An indicator tuned on healthy volunteers cannot be assumed to read a damaged or non-standard substrate correctly. That is an argument for the calibration-first approach the project’s consoles take, where any indicator must be validated on the substrate it will be used to read, before it is used to make claims about emerging consciousness in that substrate.

The severe-AD caveat carries its own lesson for artificial systems. Limited fronto-parietal markers in minimally responsive patients may mean absent awareness or an unresponsive measurement channel. Distinguishing those two is exactly the problem an AI consciousness assessment faces when the subject is a system that cannot talk.

What stays open

The preprint is cross-sectional and the severe-AD subgroup was behaviourally minimally responsive, so the null-ish fronto-parietal result cannot separate lost awareness from lost task engagement. The authors name that limit directly. Longitudinal follow-up, and indicators that do not depend on task compliance, are the stated next steps. The study also leaves the connection between awareness and the metacognitive layer untested in severe patients, which is where the wider 2026 wave of consciousness-measurement work is converging. What the preprint establishes is that the awareness signature degrades smoothly with neurodegeneration, which makes the dementia clinic the most sensitive natural laboratory available for testing whether a consciousness indicator actually tracks the thing it claims to read.

Researchers covered here

• JH Jonathan HuntleyUniversity of Exeter; Division of Psychiatry, University College LondonAwareness assessment in severe Alzheimer's disease, multimodal EEG-fMRI measures of perceptual awareness 1 article
• PM Pedro MedianoImperial College London and the University of CambridgeIntegrated Information Decomposition, and the revised non-negative measure of integrated information 4 articles
• SF Stephen M. FlemingUniversity College LondonMetacognition research and the sensory horizons account of conscious vision 2 articles
• DB Daniel BorDepartment of Psychology, University of Cambridge; Centre for Brain and Behaviour, Queen Mary University of LondonPrefrontal-parietal network studies of consciousness, working memory and chunking, the rule-plus-exception model of categorization 1 article