Brain Anatomy

35 brain structures mapped through the GCD kernel. Every region measured across 8 channels — volumetric fraction, segmentation accuracy, multimodal contrast, structural heterogeneity, bilateral symmetry, clinical sensitivity, connectivity degree, and resolution dependence. Computed live in your browser.

Quid supersit post collapsum cerebri? — What survives after the collapse of the brain?

HoliAtlas 2026 NextBrain 2025 35 Structures 8 Channels 7 Categories 6 Theorems Proven

§1 — Interactive Brain Model

Click any brain region to see its GCD kernel analysis. The model shows a sagittal (side) view with major structures labeled and color-coded by category.

Frontal Parietal Occipital Temporal Insula Thalamus Caudate Put. GP STN CA1 CA2/3 Amyg Cingulate Midbrain Pons Medulla SN RN Ant. Cereb. Post. Cereb. Vermis Corpus Callosum Fornix Hypo Pulv LGN MGN
Cortical
Subcortical
Limbic
Brainstem
Cerebellar
White Matter
Specialized

Click a brain region to see its kernel analysis

35 structures across 7 categories

§2 — Kernel Atlas: All 35 Structures

Every structure measured through K: [0,1]⁸ × Δ⁸ → (F, ω, S, C, κ, IC). Sort by any column. Hover for channel breakdown.

Structure ↕ Category ↕ F ↕ ω ↕ IC ↕ Δ ↕ S ↕ C ↕ Regime

§3 — Neural Pathways: How the Brain is Wired

Sensory Input Pathways

Visual Pathway

Retina LGN V1 (Occipital) V2/V4 Temporal (what) / Parietal (where)

The lateral geniculate nucleus (LGN) is the thalamic relay for vision. Its resolution_dependence = 0.90 — one of the highest in the atlas, because its 6 layers are only resolvable at ultra-high resolution.

Auditory Pathway

Cochlea Brainstem nuclei MGN A1 (Temporal) Wernicke's area

The medial geniculate nucleus (MGN) relays auditory information to cortex. Like LGN, it has very high resolution_dependence (0.90) and very low volumetric_fraction (0.03) — classic geometric slaughter pattern.

Motor Pathway

Prefrontal (plan) Motor cortex Corticospinal tract Spinal cord

Parallel loop: Cortex → Putamen → GP → Thalamus → Cortex (basal ganglia circuit). Dopaminergic modulation from substantia nigra. Parkinson's disrupts this loop.

Memory Circuit (Papez)

Hippocampus Fornix Mammillary body Mammillothalamic tract Anterior thalamus Cingulate

The Papez circuit is the anatomical substrate of memory consolidation. Every structure in this loop has high resolution_dependence — the circuit is only fully visible at ultra-high resolution.

Connectivity Hubs

The thalamus and corpus callosum are the brain's primary structural hubs — nearly all information passes through one or both. Connectivity_degree channel values reflect their hub weight from HCP diffusion tractography.

§4 — How Thought Works

A thought is not a thing — it is a collapse-return cycle across distributed neural networks. The brain continuously collapses sensory input into compressed representations, and those representations return through recursive processing to become conscious experience. The GCD kernel measures what survives each stage of this cycle.

The Anatomy of a Single Thought

👁️
1. Sensation
Raw sensory input collapses 10⁹ bits/sec to ~50 bits/sec of conscious awareness. The collapse ratio is 20 million:1.
F ≈ 0.000003%
2. Thalamic Relay
The thalamus (F = 0.782, highest in atlas) filters, amplifies, and routes signals. It is the brain's central switchboard — 13 nuclei per hemisphere.
F = 0.782, IC = 0.650
🔄
3. Cortical Processing
Recurrent loops between cortical areas (frontal ↔ parietal ↔ temporal) integrate features into a coherent representation. This takes ~300ms.
⟨F⟩_cortical = 0.670
🧠
4. Global Workspace
The "winning" representation is broadcast globally via long-range white matter tracts (corpus callosum, arcuate fasciculus). This broadcast IS consciousness.
Broadcast F > threshold
💭
5. Return
The thought returns to hippocampal memory circuits for encoding. If τ_R is finite, the thought is real (it returns). If τ_R = ∞_rec, it was a gesture — a thought that didn't stick.
τ_R: finite → memory

The Binding Problem & Gamma Oscillations

How does the brain bind color, shape, motion, and meaning into a single unified percept? The answer involves gamma oscillations (30–100 Hz) — fast electrical waves that synchronize distributed neural populations.

In GCD kernel terms: gamma oscillations are the mechanism that maintains multiplicative coherence (IC) across channels. When gamma synchrony breaks down, IC drops (geometric slaughter) and the unified percept dissolves — exactly what happens in anesthesia, where IC/F drops from ~0.85 to ~0.20.

Neural Correlates — Kernel Analysis
Gamma oscillation: temporal_integration = 0.95, highest binding NCC
Global workspace: highest overall F among all NCCs
Thalamic gating: lowest category F — narrow function, high selectivity
Cross-frequency coupling: bridges slow (theta) and fast (gamma) rhythms
Binding = Multiplicative Coherence

The binding problem IS the heterogeneity gap problem. If one channel (e.g., color processing) desynchronizes while others remain active, IC drops catastrophically even though F stays high.

This is geometric slaughter applied to neural processing:

IC = exp(Σ wᵢ ln(cᵢ)) → one cᵢ ≈ 0 kills IC

A patient with intact but unsynchronized cortical processing has high F (individual channels working) but low IC (no binding) — precisely the signature of disorders of consciousness.

The heterogeneity gap Δ = F − IC measures binding failure.

Modes of Thought — Kernel Signatures

Focused Attention
Prefrontal cortex suppresses competing stimuli. High F in task-relevant channels, low C (curvature). The brain narrows its trace vector.
High F · Low C · Low S → Stable approach
Mind-Wandering (DMN)
Default Mode Network activates during unfocused states. Moderate F but high S (entropy) — the brain explores its own state space. Creativity emerges from high-entropy exploration plus return.
Moderate F · High S · High C → Watch regime
Flow State
Optimal performance: prefrontal activity decreases (transient hypofrontality), motor and sensory cortices dominate. IC/F approaches 1.0 — extremely homogeneous channel activation across the engaged network.
High F · High IC/F · Low Δ → Near-Stable

Working Memory — The 4±1 Item Limit

Working memory holds ~4 items simultaneously. Why this limit? The prefrontal cortex maintains active representations through persistent firing (delay activity). Each additional item adds a channel to the trace vector. As channels multiply, maintaining multiplicative coherence (IC) becomes exponentially harder — adding a 5th item while keeping 4 others active causes IC to drop below the conscious threshold.

1 item
2 items
4 items
7 items
IC/F ratio drops as items increase — the working memory limit IS geometric slaughter

§5 — Awareness & Consciousness

Awareness is measured, not assumed. The GCD kernel quantifies consciousness through two complementary approaches: the awareness-cognition kernel (34 organisms across phylogeny, 10 channels) and the coherence kernel (20 systems, 8 channels measuring harmonic, recursive, and phase properties).

The Consciousness Spectrum

Consciousness is not binary. The kernel reveals a continuous spectrum from minimal awareness (C. elegans, 302 neurons) to recursive self-modeling (human adult). The key insight: no organism reaches Stable regime — consciousness is inherently a Watch or Collapse phenomenon. Stability would mean static awareness, which contradicts the recursive, self-updating nature of consciousness.

← Less awareness Greater awareness →

The 5+5 Awareness-Aptitude Structure

Awareness (Reflective) — 5 channels

Mirror recognition Self-model capacity
Metacognitive accuracy Calibration of confidence
Planning horizon Temporal depth of future
Symbolic depth Recursive language/symbol
Social cognition Theory of mind
Human adult: (0.95, 0.85, 0.90, 0.95, 0.88) → Aw ≈ 0.906

Aptitude (Somatic) — 5 channels

Sensory acuity Species-optimal discrimination
Motor precision Fine motor control
Environmental tolerance Survivable conditions range
Reproductive output Fecundity × viability
Somatic resilience Recovery, longevity, stress
Human adult: (0.72, 0.80, 0.55, 0.35, 0.70) → Ap ≈ 0.624

The Awareness-Aptitude Trade-off

Across 34 organisms, awareness and aptitude are anti-correlated (ρ < −0.50). High cognitive investment comes at somatic cost. The human adult has Aw = 0.906 but Ap = 0.624 — extreme awareness, moderate somatic fitness. Compare: mantis shrimp has Aw = 0.10 but Ap = 0.92 — minimal awareness, extreme somatic competence. This trade-off is the heterogeneity gap in the awareness kernel: IC = √(Aw·Ap) is always less than F = (Aw+Ap)/2.

Brain Structures Critical for Consciousness

Thalamo-Cortical Loop

The thalamus (F = 0.782) doesn't just relay — it gates consciousness. The reticular nucleus wraps the thalamus like a shell, selectively inhibiting relay neurons. When the reticular gate opens, sensory information reaches cortex and becomes conscious. When it closes (as in deep sleep), consciousness fades.

This is why anesthesia targets thalamic circuits first — disrupting the gate disrupts consciousness itself, regardless of cortical integrity.

Hippocampal Memory Binding

Consciousness requires temporal continuity — you can't be conscious of "now" without reference to the immediate past. The hippocampus (CA1: F = 0.740) provides this by binding current experience to recent memory. CA1 is the first subfield lost in Alzheimer's — and the first aspect of consciousness to degrade.

Clinical sensitivity = 0.95 (highest in atlas). Resolution_dependence = 0.92. CA1 subfield segmentation requires ultra-high resolution to detect early changes.

Prefrontal Cortex — The Executive

The prefrontal cortex (F = 0.730) is the seat of executive function: working memory, planning, decision-making, and self-regulation. It is the last region to myelinate (age ~25) and the first to show age-related decline. Its connectivity_degree = 0.90 (second only to thalamus) reflects its role as an integration hub.

The "cognitive executive" is not a homunculus — it's the region with the most diverse connections, allowing it to modulate nearly every other brain area.

Claustrum & Insular Cortex

The insular cortex (F = 0.611) processes interoception — awareness of the body's internal states (heartbeat, breathing, gut feelings). Damage to the insula disrupts bodily self-awareness. Francis Crick proposed the nearby claustrum as the "conductor" of consciousness — it connects to every cortical area and may coordinate binding.

The insula's relatively low F reflects its deep, hidden location and small volume — geometric slaughter in action for consciousness infrastructure.

§6 — Attention: The Selection Engine

Attention is the mechanism that selects which information enters the global workspace (and thus consciousness). 12 attention mechanisms analyzed through the GCD kernel. The key finding: divided attention mechanisms are in Collapse regime — the brain cannot truly multitask.

Selective Attention

Focus on one stimulus while suppressing others.

• Visual selective: highest spatial resolution (0.95)

• Auditory selective: highest temporal resolution (0.90)

• Cross-modal: binds across senses

Regime: Watch

Sustained Attention

Maintaining focus over extended periods.

• Vigilance: high fatigue resistance (0.75)

• Executive: high cognitive load (0.85)

• Monitoring: moderate automation

Regime: Watch

Divided Attention

Attempting to process multiple streams.

• Dual-task: capacity 0.30 → COLLAPSE

• Task-switching: most trainable (0.90)

• Time-sharing: moderate capacity

Regime: Collapse

Orienting

Directing attention to novel stimuli.

• Reflexive: highest F of all mechanisms

• Voluntary: deliberate redirection

• Inhibition of return: prevents re-fixation

Regime: Watch

Why you can't truly multitask: Dual-task attention has capacity_limit = 0.30, which means ω ≥ 0.30 → automatic Collapse regime. The brain doesn't simultaneously process two tasks — it rapidly switches between them, losing coherence (IC) at each switch. Task-switching has the highest trainability (0.90) but currently the weakest F (≈ 0.42), suggesting it can be improved but starts from a position of structural weakness.

§7 — Disease Mapping: What Ultra-High Resolution Reveals

Alzheimer's Disease

Primary targets: Hippocampus CA1 (F=0.740), CA2/3, amygdala lateral & basal

Secondary: Temporal cortex, cingulate cortex, fornix

Key finding: CA1 subfield neurodegeneration is the earliest detectable marker. Standard 1mm³ MRI cannot resolve CA1 from CA2/3. Ultra-high resolution (0.125 mm³) resolves 7 hippocampal subfields, enabling 35% detection gain over standard MRI.

Detection gain from subfields: +35%

Parkinson's Disease

Primary targets: Substantia nigra (F=0.680), red nucleus, subthalamic nucleus

Secondary: Putamen, caudate, globus pallidus

Key finding: Deep gray matter nuclei volume loss precedes motor symptoms by years. These nuclei are only identifiable with ultra-high resolution — pBrain achieves Dice = 0.89 for deep GM.

Detection gain from subfields: +40%

Frontotemporal Dementia

Primary targets: Frontal cortex (F=0.730), temporal cortex, insular cortex

Secondary: Amygdala lateral, cingulate cortex

Key finding: MRI-based anatomical staging (Planche 2023) can track disease progression through cortical volume patterns. Large cortical structures have lower resolution_dependence — FTD is more detectable with standard resolution than other dementias.

Detection gain from subfields: +25%

Multiple Sclerosis

Primary targets: Corpus callosum (F=0.638), corticospinal tract, cerebellar WM

Secondary: Arcuate fasciculus, uncinate fasciculus, pons

Key finding: WMn (White Matter nulled) modality specifically enhances white matter lesion detection. Standard T1w/T2w miss many periventricular lesions. This is HoliAtlas's unique advantage — first atlas with synthetic WMn.

Detection gain from subfields: +30%

§8 — Brain Atlas Theorems (6/6 Proven)

§9 — Brain Atlas Landscape: 30 Years of Progress

From MNI152 (1993, 1mm³, ~100 labels) to HoliAtlas (2026, 0.125mm³, 350 labels) — a factor of 8× in resolution and 3.5× in label density. Each atlas represents a different generation of brain mapping technology.

Atlas Year Resolution (mm³) Labels Modalities vs HoliAtlas (res.)

HoliAtlas Segmentation Pipeline — 7 Software Packages Fused

315 total regions. Mean Dice coefficient across all packages: 0.8337. Each package specializes in different brain structures.

§10 — Data Sources

HoliAtlas (2026)

Authors: Manjón, Morell-Ortega, Ruiz-Perez et al. (19 authors)

Journal: Scientific Reports 16:9457

DOI: 10.1038/s41598-026-40186-2

Resolution: 0.5×0.5×0.5 mm (0.125 mm³)

Matrix: 362 × 434 × 362 voxels

Modalities: T1w, T2w, WMn (synthetic White Matter nulled)

Subjects: 75 from HCP1200 (41F, 34M, ages 22–35)

Labels: 350 substructures → 54 structures → 9 tissues → 1 ICV

Scanner: 3T MR

Construction: 3 years

NextBrain (2025)

Lead: Juan Eugenio Iglesias (UCL / MGH Harvard)

Journal: Nature

DOI: 10.1038/s41586-025-09708-2

Resolution: 100 μm (0.001 mm³)

Regions: 333

Method: AI-assisted probabilistic histological atlas

Brains: 5 post-mortem × 10,000 sections each

Validation: 3,000+ living MRI scans

Platform: FreeSurfer

Construction: 6 years