# Neuron — The Basic Unit of the Nervous System > *The neuron is the basic structural and functional unit of the nervous system. About 86 billion neurons (Azevedo 2009) form the human brain, interconnected by 100 trillion synapses. This article covers neuron anatomy, main types (pyramidal / interneuron / Purkinje), basic electrophysiology, comparison with modern deep learning.* > > **Difficulty**: Introduction-Intermediate > **Prerequisites**: [Neuroscience History](/en/mindresearch/neuroscience/00_Foundations/Neuroscience_History) --- ## 1. Neuron Basic Anatomy

Figure 1 — Typical pyramidal neuron structure: dendrites collect many inputs, soma integrates, axon outputs spikes, synapses pass to next neuron.

### 1.1 Soma (Cell Body) - Nucleus + cytoplasm - Diameter 4-100 μm - Protein synthesis center - Integrates dendrite inputs, decides whether to spike ### 1.2 Dendrites - Receive inputs (10³-10⁴ synapses per pyramidal neuron) - Active (contain voltage-gated channels) — nonlinear integration - Branching pattern defines neuron type ### 1.3 Axon - Output end, up to 1m+ (motor neuron) - Myelin sheath: formed by Schwann cells (PNS) / oligodendrocytes (CNS) - Myelin accelerates signal (saltatory conduction) - Ranvier nodes are unmyelinated segments, concentrate Na+ channels ### 1.4 Axon Terminal (Pre-synaptic Terminal) - Contains vesicles (loaded with neurotransmitter) - Action potential arrives → Ca²⁺ inflow → vesicle fuses, releases NT --- ## 2. Resting Membrane Potential Neuron at rest: -70 mV intracellular (relative to outside): - High intracellular K⁺ (140 mM) - High extracellular Na⁺ (145 mM) + Cl⁻ - Na⁺/K⁺ pump (ATPase) maintains gradient: 3 Na⁺ out + 2 K⁺ in per cycle - At rest mainly K⁺ leak channels open → close to K⁺ equilibrium potential (-90 mV) Nernst equation: $$E_K = \frac{RT}{zF} \ln \frac{[K^+]_o}{[K^+]_i}$$ --- ## 3. Action Potential Classic 1ms event: 1. **Depolarization**: threshold ~ -55 mV, Na⁺ channels open 2. **Peak**: ~+30 mV (Na⁺ inflow) 3. **Repolarization**: Na⁺ channels inactivate, K⁺ channels open, K⁺ outflow 4. **Hyperpolarization**: K⁺ channels slowly close 5. **Return to rest**: Na⁺/K⁺ pump All-or-none: consistent spike above threshold; no fire below. --- ## 4. Main Neuron Types ### 4.1 Pyramidal Neurons - Main excitatory neurons of cortex - Triangular soma + apical dendrite (branches upward) - ~70% of cortex - Release glutamate ### 4.2 Interneurons - Mainly inhibitory (GABA) - Types: - **PV (Parvalbumin)**: fast-spiking, control network timing - **SOM (Somatostatin)**: modulate dendrite - **VIP**: disinhibition - ~20-30% of cortex ### 4.3 Purkinje Neurons - Cerebellum-specific - **Massive dendrite tree** (~200k inputs) - Output inhibitory (GABA) ### 4.4 Dopamine Neurons (VTA/SNc midbrain) - ~600k in humans - Control reward / motor - Parkinson's = SNc degeneration ### 4.5 Granule Cells - Cerebellum + hippocampal dentate gyrus - Most abundant type --- ## 5. Biological vs AI Neural Network | Aspect | Biological neuron | AI neuron | |---|---|---| | Output | spike (binary, 0-1ms) | continuous (float) | | Integration | dendrite nonlinear + soma | weighted sum | | Time | precise spike timing | synchronous forward pass | | Learning | STDP, neuromodulation | Backprop, SGD | | Count | 8×10¹⁰ | GPT-4 ~ 10¹² params | | Connectivity | sparse, local + long | dense matrix | | Energy | 20W brain | 10kW GPU | → AI vastly **simpler** than brain; brain at least 1-2 orders more complex. --- ## 6. PyTorch — Simplified Neuron Model ```python import torch class LIFNeuron: """Leaky Integrate-and-Fire — simplest spiking neuron.""" def __init__(self, V_rest=-70, V_th=-55, V_reset=-75, tau=20): self.V = V_rest self.V_rest, self.V_th, self.V_reset, self.tau = V_rest, V_th, V_reset, tau def step(self, I_input, dt=0.1): dV = (self.V_rest - self.V + I_input) / self.tau * dt self.V += dV if self.V >= self.V_th: self.V = self.V_reset return 1 return 0 neuron = LIFNeuron() spikes = [] for t in range(1000): I = 12 if 200 < t < 800 else 0 spikes.append(neuron.step(I)) ``` --- ## 7. Count and Connectivity Human brain: - 86 billion (8.6 × 10¹⁰) neurons (Azevedo 2009) - 100-500 trillion synapses - Density varies regionally: cerebellum 80% of neurons but 10% of volume - Total axon length ~165,000 km (4× around Earth) --- ## 8. Relation to Glia Non-neuronal cells (glia): - Astrocytes: nutritional support, BBB - Oligodendrocytes: CNS myelin - Microglia: immunity / synaptic pruning - Schwann cells: PNS myelin Glia:neuron ≈ 1:1 (old 10:1 view corrected). --- ## 9. Neurogenesis Adult brain can still produce new neurons: - Hippocampus dentate gyrus (confirmed) - Olfactory bulb (rodents) - Cortex (disputed) → Stem cell-based therapies explored in Parkinson etc. --- ## 10. Common Pitfalls ### 10.1 "Neuron Count Determines Intelligence" Misconception Humans (8.6 × 10¹⁰) vs elephants (2.5 × 10¹¹) — elephants have more. But humans have more cortex neurons + higher connectivity. ### 10.2 LIF ≠ Real Neuron LIF is a toy model; real neurons have dendritic computation + neuromodulation etc. ### 10.3 Hodgkin-Huxley Axon Only HH measured in axon; dendrite dynamics differ (NMDA etc. complex). ### 10.4 Spike Timing Matters Many ML treats firing rate as output, ignores timing. Real brain heavily depends on precise timing. ### 10.5 Glia Aren't Just "Support" Past 20 years revealed glia participate in synaptic plasticity / signaling. --- ## 11. Related Concepts - **History**: [Neuroscience History](/en/mindresearch/neuroscience/00_Foundations/Neuroscience_History) - **AI comparison**: https://jeffliulab.github.io/ai-notes/02_Deep_Learning/01_Intro/MLP/ --- ## References 1. **Kandel, E. R. et al.** *Principles of Neural Science*. 6th ed., 2021. 2. **Azevedo, F. et al.** "Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain." *J Comp Neurol*, 2009. 3. **Hodgkin, A. L. & Huxley, A. F.** *J Physiol*, 1952. 4. **Cajal, S. R.** *Histology of the Nervous System*. 1909-1911. 5. **Bear, M. F. et al.** *Neuroscience: Exploring the Brain*. 4th ed., 2015.