A study mapped 10.4 million brain cells and found between 25 and 1,300 distinct neighborhoods, depending on the resolution chosen.
This is not imprecision. It is the answer. The brain does not have a fixed number of regions. It has a continuous hierarchy of structural organization, and the number of “regions” you find depends entirely on the granularity at which you look. At coarse resolution, the cortex divides into a few dozen areas. At fine resolution, each of those areas fractures into hundreds of sub-neighborhoods with distinct cell-type compositions and connectivity patterns.
The practical consequence resolved a genuine scientific conflict. Previous studies had assigned contradictory functions to what appeared to be the same brain region. One group would localize a function to area X; another group would show area X doing something entirely different. The CellTransformer analysis revealed that “area X” was not one area. At the resolution those studies operated, multiple neighborhoods with distinct cell compositions and presumably distinct functions were being collapsed into a single label. The studies weren't contradicting each other. They were sampling different neighborhoods of a region that doesn't exist as a unit.
This is a scale-dependent observation problem. The brain has structure at every level of description, and no level is privileged. Brodmann's 52 areas are not wrong — they describe real boundaries at one resolution. The Allen Brain Atlas's finer parcellation is not more correct — it describes real boundaries at another resolution. The 1,300-neighborhood decomposition is not the final answer — a still finer analysis would find structure within those neighborhoods. The question “how many brain regions are there?” has no answer, the way “how long is the coastline of Britain?” has no answer. The measurement creates the count.
The tool that revealed this — CellTransformer, a deep learning model trained on spatial gene expression — is itself interesting for what it implies about discovery. The hierarchical neighborhood structure was always present in the data. Earlier analyses couldn't find it because they lacked a method for simultaneously representing structure at multiple scales. The biological reality didn't change. The analytical resolution did. The discovery is real, but what was discovered is that the previous descriptions were incomplete, not that they were wrong.