Repository navigation
[REF] Flip the bits on the histological profiles #274
Description
Activity
Still relevant. The current
read_histology_profilereturns the raw 8-bit values from BigBrainWarp, where lower = darker = more cell-dense — opposite to the convention used in Paquola et al. 2021 and most downstream BigBrain analyses.Two ways forward:
- Flip by default and add an
invert=Falseescape hatch for users who depend on the current behaviour. This is the change the issue calls for and is the cleanest. The bit-flip itself is a one-liner:profiles = np.iinfo(profiles.dtype).max - profilesfor the 8-bit array (or255 - profilesif we keep it explicit). MPC (compute_mpc) is correlation-based, so partial correlations and gradients are unaffected, as the issue notes — but downstream code that inspects the raw profile values directly would see a flip, hence the toggle. - Document only, no behaviour change. Safer for downstream reproducibility but leaves the "intuitive" convention out of reach by default.
Option (1) seems preferable, paired with a clear note in the changelog and the
read_histology_profiledocstring. If anyone has objections — particularly if you have published pipelines that pin against current BrainStat output — please weigh in here before someone opens a PR.- Flip by default and add an
Implemented in #380.
read_histology_profilenow flips the BigBrainWarp 8-bit values so higher = darker = denser (matching Paquola et al. 2021), gated oninvert: bool = True. Passinvert=Falseto keep the previous output. MPC and gradient computations are correlation-based, so they're unaffected by the flip — only direct intensity inspection changes.- added a commit that references this issue
on May 4, 2026
For the histological module "The Merker staining labels cell bodies, similar to Nissl staining, with a high contrast between black cell bodies on a light background (Merker, 1983). In the digitised images, darker colouring is represented by lower numbers (8bit graphics: 0–28 = black-white). It is common practice to invert the values of the intensity, such that image intensity increases with staining intensity." Paquola et al., 2021, eLife
Currently, these values are not inverted in BrainStat, we should add this as a default.
Note that this should not affect MPC computations.