Brodmann Areas: Cytoarchitectonic Atlas

• Brodmann areas are defined regions of the cerebral cortex distinguished by cytoarchitectonic features such as cell density, layer thickness, and cellular composition.
• They are mapped through classical histological methods, fMRI, and modern CNN-based segmentation, facilitating precise functional localization and brain mapping.
• Despite widespread use, limitations such as inter-individual variability and static parcel definitions drive the development of alternative, dynamic parcellation frameworks.

Brodmann areas are anatomically defined regions of the primate cerebral cortex demarcated by their cytoarchitecture—variations in the size, density, and organization of neuronal cells across the cortical laminae. Introduced in 1909 by Korbinian Brodmann through systematic mapping of Nissl-stained postmortem brains, this parcellation assigns discrete numbers (BA1–52) to morphologically distinct patches of cortex. Brodmann’s scheme has provided a foundational spatial reference for functional localization, brain mapping, and comparative studies, underpinning both traditional neuroanatomy and modern neuroimaging methodologies. Despite their widespread usage, Brodmann areas represent an idealization that may not consistently capture the full functional or connectivity-based organization of cortex due to intrinsic biological variability and methodological limitations.

1. Historical Origins and Cytoarchitectonic Criteria

Brodmann’s "Vergleichende Lokalisationslehre der Grosshirnrinde" (1909) established the first comprehensive atlas of the primate neocortex based on differences in microscopic anatomy visible in stained tissue sections. Regions were distinguished by:

• Cell-packing density: Quantitative differences in the relative density of neurons across layers I–VI.
• Layer thickness: For example, the presence of a thick granular layer IV in primary sensory cortices, versus a thinner or absent layer in agranular motor areas.
• Cellular composition: Distribution and size of pyramidal cells in layers III and V, or the predominance of specific cell types.
• Transition sharpness: Abrupt versus gradual changes between adjacent regions.
• Granularity: Variation between granular (e.g., primary sensory) and agranular (e.g., motor) cortex.

Modern interpretations use normalized cell-density profiles $C(z)$ as a function of cortical depth $z$, with quantitative borders sometimes placed using Mahalanobis distance between adjacent profiles. Brodmann’s numbering—BA1 through BA52 in humans, with slight differences in non-human primates—remains the standard reference system, though not every area is present or consistently delineated in all individuals (Hayden, 2022).

2. Canonical Brodmann Areas: Locations and Functions

Brodmann’s cytoarchitectonic map associates discrete numbers with reproducible anatomical loci, most with established functional correlates. While not exhaustive, the following summary conveys the breadth of the scheme:

Brodmann Area(s)	Anatomical Location / Function
BA1–3	Primary somatosensory cortex (postcentral gyrus)
BA4	Primary motor cortex (precentral gyrus)
BA5–7	Somatosensory and visuomotor association (superior parietal)
BA6	Premotor and supplementary motor area
BA8	Frontal eye fields
BA9–12, 46, 47	Dorsolateral/ventrolateral prefrontal cortex (executive, working memory)
BA10	Frontopolar cortex (abstract cognition, multitasking)
BA17	Primary visual cortex (V1)
BA18–19	Visual association areas (V2–V4)
BA20–21	Inferotemporal cortex (object recognition)
BA22	Superior temporal gyrus (Wernicke’s language area)
BA23–26, 29–31	Cingulate cortex (emotion, attention, default mode)
BA27–28	Parahippocampal cortex (memory)
BA30–36	Perirhinal/entorhinal cortex
BA37	Fusiform gyrus (face/place processing)
BA39–40	Inferior parietal lobule (language, spatial attention)
BA41–42	Primary/secondary auditory cortex
BA44–45	Broca’s speech area
...	...

The mapping between area, histology, and function has been instrumental in standardizing research on cortical localization of function (Hayden, 2022).

3. Methodologies for Identification and Mapping

Histological and Microscopy-Based Mapping

Classically, Brodmann areas are delineated via postmortem histology, using stains (e.g., Nissl, Merker) that reveal laminar architecture. Manual annotation is augmented by observer-independent metrics (e.g., GLI profiles), sometimes with digital quantification of laminae and borders. For large-scale datasets, new workflows employ convolutional neural networks (CNNs) to automate cytoarchitectonic border segmentation (Schiffer et al., 2020):

• Annotation strategy: Sparse manual annotation (~1% of sections) in reference slices; local CNNs trained on adjacent annotated pairs interpolate labels to densely fill gaps.
• Architecture: Multi-scale U-Net–style encoder-decoders process co-centered image patches at multiple resolutions, capturing both fine laminar texture and macroscopic folding.
• Evaluation: Networks are evaluated using held-out (unseen) sections and the F₁ (Dice) score, achieving median F₁ ≈ 0.75 for multi-scale architectures on 18 cortical areas, outperforming global models (F₁ ≈ 0.53) (Schiffer et al., 2020).
• Robustness: The method tolerates large histological artefacts and works efficiently with terabyte-scale data.

Functional MRI and Projective Approaches

Functional neuroimaging routinely maps fMRI signals to Brodmann areas using standardized atlases and coordinate systems (e.g., Talairach space):

• Projection: Each voxel’s (x, y, z) coordinate is assigned to a Brodmann area based on atlas boundaries; signals are then averaged within each area.
• Correlation and PCA: The Pearson correlation matrix $C_{ij}$ is computed across Brodmann areas, followed by principal component analysis (PCA) of $C$ to extract collective modes. Eigenvalues above the Marchenko–Pastur threshold (from random matrix theory; $λ_+ ≈ 2.25$ for $N=41, T=163$) signify true functional collectivity, while the bulk is attributed to noise (Burda et al., 2013).
• Participation Ratio (PR): The degree of spatial distribution for each mode is quantified by the PR, with PR ≈ 1 indicating localization to few areas, PR ≈ 41 indicating widespread involvement.
• Example Result: In combined idle and tapping states, the leading mode (λ₁=22.627, PR=34.39) corresponds to cortex-wide coactivation. The second mode (λ₂=4.478, PR=16.42) localizes to prefrontal/frontal areas associated with motor planning, with its distribution shifting between rest and task (Burda et al., 2013).

4. Interpretive Limitations and Controversies

Despite broad utility, multiple limitations of Brodmann areas are identified (Hayden, 2022):

• Parcel size mismatch: Some areas are too large—encompassing multiple functional micro-domains not reflected in the original borders. Others are too small to encapsulate distributed functional processes.
• Inter-individual variability: Significant differences exist in area border placement between individuals, even after spatial normalization (e.g., MNI space), leading to misalignment between atlas-based and true histological borders.
• Ambiguous landmarks: Gyral patterns such as the H-shaped sulcus for Broca’s area (BA44/45) vary, shifting cytoarchitectonic boundaries across individuals and studies.
• Circularity and reification: There is a tendency for function to be ascribed and repeatedly confirmed within a given area, potentially obscuring distributed or network-based processes.
• Neglect of supra-areal networks: Functional networks (e.g., default mode network) and domains often cross area boundaries, contrary to a strictly areal model.

Hayden argues that an exclusive focus on Brodmann areas introduces a structure-first bias, potentially impeding the recognition of dynamic, distributed computation spanning classical parcels (Hayden, 2022).

5. Alternative Parcellation Frameworks

Newer parcellation approaches challenge or supplement Brodmann’s cytoarchitectonic scheme:

• Connectivity-based parcellations: Diffusion MRI and structural or functional connectivity groupings segment cortex by projection patterns and time-series correlation, yielding finer or coarser subdivisions than Brodmann areas (Hayden, 2022).
• Population-dynamics models: These frameworks focus on neuronal ensemble activity, identifying low-dimensional manifolds that transcend anatomical area boundaries.
• Distributed-consensus/emergentist models: Inspired by collective intelligence, these treat the cortex as an entangled system of reconfiguring modules, emphasizing fast-timescale reorganization over fixed parcellation.
• Artificial neural network informed perspectives: Deep learning models suggest that functional specialization can arise from distributed activation patterns, not rigid parcels.

A plausible implication is that while Brodmann areas remain useful as a spatial referent, modern neuroscience increasingly integrates areal, network, and dynamic frameworks to better capture brain function (Hayden, 2022).

6. Computational Advances in Brodmann Mapping

The adoption of deep CNNs for cytoarchitectonic mapping in massive histological datasets marks a significant methodological advance (Schiffer et al., 2020):

• Workflow efficiency: Localized U-Net models can infer hundreds of high-resolution intermediate section labels from minimal expert annotation, operating robustly on raw 2D sections and tolerating common artefacts.
• Scalability: Patch-based and distributed GPU implementations facilitate the processing of terabyte-scale image series, and browser-based interfaces democratize access to these pipelines.
• Interoperability: Combining CNN-mapped areas with existing 2D/3D reconstruction tools aligns high-precision microstructural maps with classical Brodmann definitions at unprecedented scale and accuracy.

This suggests the future of Brodmann mapping will likely fuse data-driven CNN-based segmentations with classical cytoarchitectonic and functional reference systems, improving reproducibility and resolution across research paradigms (Schiffer et al., 2020).

7. Significance and Outlook

Brodmann areas have shaped neuroscience’s conceptualization of cortical organization but must be considered in the context of methodological limitations and advances in brain mapping. While their cytoarchitectonic basis ensures a reproducible anatomical vocabulary, the complexity of functional, connectional, and dynamic architectures necessitates hybrid approaches. CNN-based workflows now permit data-rich, reproducible segmentations that closely follow both classical and emerging domain boundaries. Nevertheless, as highlighted by criticisms of areal rigidity, the entangled nature of brain computation calls for integration of Brodmann-based priors with connectivity, population-dynamic, and emergentist models, providing a pluralistic foundation for future research (Burda et al., 2013, Schiffer et al., 2020, Hayden, 2022).