Mobile cognition: imaging the human brain in the "real world"

Nature Reviews Neuroscience 24:347–362, doi 10.1038/s41583-023-00692-y; NIH author manuscript, PMC10642288 (NIHMS1935607). A Perspective (review with opinion), not a primary study. Read for research direction R1, 4 October 2026. Provenance: papers/amadeus/stangl2023_mobile_cognition.provenance.json.

What was read

• Read in full: the PMC author manuscript (NCBI efetch XML converted to text): abstract, every section (methods, the four application areas, challenges, conclusions), Box 1, all four figure captions, and all 175 references as bibliographic entries.
• Not read: the cited primary studies (references were read as entries only), and the figures as images. There is no supplement.

What it says

Methods for recording people who move (as the authors describe them):

• Mobile scalp EEG: miniaturised and wearable. Highly susceptible to motion artefact, which "advanced analysis techniques can remove a reasonable amount of". It records superficial cortex; high-density source localisation has been applied to deeper regions (thalamus, retrosplenial cortex).
• Mobile fNIRS: cortical haemodynamics, with motion-artefact correction.
• OPM-MEG: optically pumped magnetometers worn like a helmet. Covers cortical and some subcortical regions, but "still dependent on 'magnetic shielding'", meaning a specially shielded room.
• Chronic closed-loop implants (responsive neurostimulation and adaptive DBS):
  • Use: "implanted in thousands of individuals", mostly for epilepsy and Parkinson's disease, more recently under investigational exemptions for depression, PTSD, OCD, binge eating and cervical dystonia.
  • Advantages: not externally visible, no obvious restriction on movement; motion-artefact-free recordings from deep regions (hippocampus, entorhinal cortex, amygdala, nucleus accumbens) "over long time periods (months or years)", plus stimulation for causal tests.
  • Coverage: "always a limited number of simultaneous recording sites (usually 4) per person", placed by clinical indication, most often in the medial temporal lobe and basal ganglia.
  • Higher channel counts: over 100 channels, including single neurons, are possible only in hospital epilepsy monitoring.
  • Epileptic activity: in previous studies "~3% of the data during experimental sessions was affected by epileptic activity".
  • Platform: the Mo-DBRS platform gives a real-time readout through a wand held over the implant, mounted on a backpack.

What mobile recordings have shown (cited, not verified here):

• Hippocampal theta during real walking: medial temporal theta at higher frequencies than in virtual reality, in bouts, more during fast walking.
• Boundaries, self and others: theta increased near room boundaries, for one's own position and for another person's, only when the location mattered for the task.
• Mobile EEG: subsequent-memory effects outdoors.
• Hyperscanning: EEG synchrony between classmates over a semester tracked engagement.
• Motor symptoms: at-home Parkinson's recordings (Gilron 2021, read in this folder).
• OCD: over 1,000 hours of at-home intracranial data, heart rate and symptom-intensity ratings. Ventral capsule/ventral striatum delta power in the minute before and after each self-report correlated negatively with self-rated symptom intensity over 3 days of continuous recording (Provenza 2021, Fig. 4c).
• Depression: symptom-specific activity in one person that later triggered closed-loop stimulation (Scangos 2021).

Labels and context streams (Box 1): optical motion capture (indoors), body-worn inertial units, body-mounted audio and video, mobile eye tracking, heart rate, respiration, skin conductance, and smartphone apps for symptom ratings and sleep quality. All must be time-synchronised with the neural data, by marker signals, timestamps or network time. Open code exists for the implant platforms.

Challenges named:

• Uncontrolled real-world variables, to be recorded and modelled rather than excluded.
• Analysis by mixed-effects and multimodal models and computational ethology.
• Sparse, clinically placed coverage of implants.
• Implant recordings exist only in patients, so disease, medication and seizures must be modelled. Non-invasive methods in healthy people are the comparison.

Limits

• A Perspective by a group that builds one of the implant platforms. Claims about findings rest on cited studies not read here, and the tone is advocacy ("a new era").
• No quantitative comparison of methods: no table of resolution, channel counts, battery life, wear time or data loss. The numbers here are the few the authors state ("usually 4" sites, about 3% epileptic data, "months or years").
• Nothing on recording thought content or autobiographical memory in daily life. The examples are navigation, social synchrony, movement and affect.

What it means for Amadeus

• The methods landscape for a healthy person is short (verified as the authors' account).
  • Mobile scalp EEG and fNIRS go anywhere but record superficial cortex with motion artefact.
  • OPM-MEG gives better localisation but only inside a shielded room.
  • Deep structures that matter for memory and affect (hippocampus, amygdala) are recorded in daily life only through clinical implants, at about four sites per person.
  • "A month under electrodes" for a healthy volunteer is therefore not on offer. The realistic invasive analogue is a patient who already has a chronic implant (inference).
• Labelling unlabelled time follows one pattern (verified). Dense automatic context streams (motion, audio, video, heart rate, location) plus sparse self-reports from a phone app. The neural window is anchored to each report (±1 minute in the OCD example). That is brief 6.7's experience-sampling design in its established form, with symptom intensity rather than thought content as the reported variable.
• Synchronisation is a design requirement, not a detail (verified). Every stream needs a common clock: marker signals, timestamps or network time protocol.
• Where to look next (pointer, not read): Provenza et al. 2021 (Nature Medicine, PMC8800455) is the closest published template for months of at-home intracranial recording combined with experience sampling and video. Topalovic et al. 2023 (Nature Neuroscience, PMC9991917) is the wearable single-neuron platform.