brain2scalp 0.1.0

Map MNI, Talairach or native-space brain coordinates to their nearest scalp entry point on NIfTI head scan

Map a brain target to its nearest scalp entry point

---

Table of Contents

• Features
• Installation
• Quick Start
• Atlas vs Personal Scans
• CLI Reference
• Python API
• Output Formats
• Visualization
• Package Layout
• License

---

Features

• Local-normal ray-cast - queries the k nearest scalp surface voxels to the brain target, takes their centroid to estimate the local outward normal, then casts a ray along that direction to find the scalp entry point. Falls back to directed nearest-neighbour if the ray misses the surface.
• MNI and Talairach input - Talairach auto-converted to MNI152 via Lancaster et al. (2007) piecewise-affine transform, with distinct left/right hemisphere matrices.
• Multiple targets - batch coordinate list in one pass.
• Multi-format output - plain text table, JSON, or CSV.
• Optional visualization - 6-panel pipeline-steps view, 3-panel orthogonal slices, and interactive Plotly 3D viewer.

---

Installation

1. Install Python

Requires Python >= 3.10.

Check if you already have it:

python --version
# or
python3 --version

If not Python 3.10.x or higher, download from https://www.python.org/downloads/

---

2. Create and activate environment (recommended)

venv

python -m venv env_name

# Activate it
# Windows:
env_name\Scripts\activate
# macOS / Linux:
source env_name/bin/activate

conda

conda create -n env_name python=3.10

# Activate it
conda activate env_name

---

3. Install brain2scalp

From PyPI (recommended):

# Core + CLI (no visualization)
pip install brain2scalp

# With visualization (matplotlib + plotly)
pip install "brain2scalp[viz]"

# Development (adds pytest, ruff, mypy)
pip install "brain2scalp[dev]"

From GitHub:

You need Git installed (git-scm.com). Then:

git clone https://github.com/LeonardoPoliti/brain2scalp.git
cd brain2scalp

# Core + CLI
pip install .

# With visualization
pip install ".[viz]"

Verify the install:

brain2scalp --help

You should see the CLI help text.

---

Requirements

Dependency	Minimum	Notes
Python	3.10
nibabel	5.0	NIfTI I/O
numpy	1.24
scipy	1.10	KD-tree, morphology
scikit-image	0.21	Convex hull filling
matplotlib	3.7	[viz] only
plotly	5.18	[viz] only

All dependencies are installed automatically by pip.

---

Quick Start

Command line examples

# Single MNI target
brain2scalp --nii mni_icbm152_t1_tal_nlin_sym_09c.nii --target -46 20 32

# Talairach target
brain2scalp --nii mni_icbm152_t1_tal_nlin_sym_09c.nii --target -46 20 32 --space talairach

# Multiple targets on one line
brain2scalp --nii head.nii --target "-46 20 32, 10 -20 45"

# Multiple targets from a CSV file
brain2scalp --nii head.nii --target-file targets.csv

# JSON output to file (format inferred from extension)
brain2scalp --nii head.nii --target -46 20 32 --output result.json

# CSV output to file
brain2scalp --nii head.nii --target -46 20 32 --output result.csv

# 3-panel orthogonal slice view of the target
brain2scalp --nii head.nii --target -46 20 32 --vizTarget

# 6-panel pipeline steps figure
brain2scalp --nii head.nii --target -46 20 32 --vizSteps

# Interactive 3D viewer
brain2scalp --nii head.nii --target -46 20 32 --viz3d

# Save all figures without displaying
brain2scalp --nii head.nii --target -46 20 32 --vizTarget --vizSteps --save-fig ./figs --no-plot

Python API examples

from brain2scalp.core import run, run_full, run_multi, run_full_multi

# Single target - lightweight result (coordinates only)
result = run(
    nii_path="mni_icbm152_t1_tal_nlin_sym_09c.nii",
    brain_target=[-46.0, 20.0, 32.0],
    is_talairach=False,
)

print(result.scalp_entry_mni)    # [x, y, z] in MNI mm
print(result.distance_mm)        # brain-to-scalp distance (mm)
print(result.projection_method)  # 'ray_cast' or 'nearest_neighbor'
print(result.to_dict())          # JSON-serializable dict

# Single target with volumetric arrays (needed for visualization)
state = run_full(
    nii_path="head.nii",
    brain_target=[-46.0, 20.0, 32.0],
    surface_thickness_mm=1.5,
    k_local=100,
    verbose=True,
)

# Multiple targets - one atlas load, one result per target
results = run_multi(
    nii_path="head.nii",
    targets=[[-46.0, 20.0, 32.0], [10.0, -20.0, 45.0]],
)

# Multiple targets with volumetric arrays
states = run_full_multi(
    nii_path="head.nii",
    targets=[[-46.0, 20.0, 32.0], [10.0, -20.0, 45.0]],
    is_talairach=False,
    surface_thickness_mm=1.5,
)

---

Atlas vs Personal Scans

Using an MNI Atlas

An MNI152 full-head atlas gives a reasonable approximation of scalp topology and is sufficient for group-level studies or when no individual scan is available.

Example atlas: mni_icbm152_t1_tal_nlin_sym_09c.nii
Download from: https://nist.mni.mcgill.ca/icbm-152-nonlinear-atlases-2009/

When to use:

• You have MNI or Talairach coordinates from a literature study or group activation map.
• No individual scan is available.
• You need a quick estimate of scalp position for target planning.

Limitations:

• Results reflect average anatomy, not the individual's head shape.
• Accuracy degrades for targets far from group-mean anatomy (e.g. atypical skull shape, lesions).

Using a Personal Head Scan

Scan requirements:

• Must be a full-head scan
• .nii or .nii.gz format with valid qform/sform header.
• 4D volumes are accepted; only the first frame is used.

What to avoid:

• Brain-extracted images (FSL BET output, FreeSurfer brain.mgz) - detected and rejected.
• Label/parcellation atlases - also detected and rejected.
• Files with missing or corrupt affine - rejected at load time.

Option A - Native space

Use this when your target coordinates come from the same scan.

Option B - Register to MNI first

Use this when your target is in MNI space but you want subject-specific scalp topology. Register the full-head scan to the MNI152 template with a linear (12-DOF) registration, then pass the registered scan with --space mni.

Note: Use a full-head MNI template as the registration target to preserve scalp geometry in the output.

---

Coordinate Space Summary

--space	Input type	Conversion applied
mni (default)	MNI152 mm coordinates	None
talairach	Talairach mm coordinates	Lancaster 2007 piecewise-affine → MNI
native	mm in the scan's own space	None (use with unregistered personal scans)

---

CLI Reference

brain2scalp --nii PATH --target COORDS [options]

Input

Flag	Default	Description
--nii PATH	required	Full-head .nii or .nii.gz file
--target COORDS	required*	Target coordinates in mm. Single: '-46 20 32'. Multiple (comma-separated): '-46 20 32, 10 -20 45'
--target-file PATH	-	CSV file with one target per row (x,y,z). Optional header auto-detected. Combined with --target; duplicates silently dropped.
--space {mni,talairach,native}	mni	Coordinate space of the input target

* Either --target or --target-file is required.

Processing

Flag	Default	Description
--threshold FLOAT	auto (Otsu)	Override head-mask intensity threshold
--surface-thickness MM	1.5	Surface shell erosion depth in mm
--k-local K	100	k-nearest scalp voxels used to estimate local ray direction

Output

Flag	Default	Description
--output PATH	stdout (text)	Save result to file. Format is inferred from extension: .txt, .json, or .csv
--quiet	off	Suppress progress messages

Visualization (requires [viz])

Flag	Default	Description
--vizSteps [IDX ...]	off	6-panel pipeline steps figure. Optionally pass target indices to limit which are shown, e.g. --vizSteps 0 2
--step-axis {0,1,2}	auto	Slice axis for --vizSteps: 0=sagittal, 1=coronal, 2=axial
--vizTarget [IDX ...]	off	3-panel orthogonal slice view centered on target + entry
--viz3d [IDX ...]	off	Interactive Plotly 3D viewer
--volume-stride N	3	Subsample every Nth voxel for the 3D point cloud. Lower = denser but slower
--save-fig DIR	display	Save all figures to this directory (created if missing)
--no-plot	off	Suppress display. Combine with --save-fig to save without showing

---

Python API

All functions share the same keyword arguments:

Parameter	Type	Default	Description
nii_path	str | Path	required	Full-head .nii or .nii.gz
brain_target / targets	list[float] / list[list[float]]	required	Target(s) in mm
threshold	float | None	None (auto)	Override head-mask threshold
is_talairach	bool	False	Convert input from Talairach to MNI
is_native	bool	False	Treat target as native scan mm space
surface_thickness_mm	float	1.5	Erosion depth for surface shell
k_local	int	100	k-nearest scalp voxels for ray direction
verbose	bool	False	Print progress to stdout

run() - single target, lightweight

from brain2scalp.core import run

result = run(
    nii_path="mni_icbm152_t1_tal_nlin_sym_09c.nii",
    brain_target=[-46.0, 20.0, 32.0],
    is_talairach=False,
    verbose=True,
)

Returns a ScalpResult:

Attribute	Type	Description
brain_target_mni	list[float]	Input target in MNI space
scalp_entry_mni	list[float]	Scalp entry point in MNI space
distance_mm	float	Euclidean distance brain→scalp
projection_method	str	'ray_cast' or 'nearest_neighbor'
coordinate_space	str	'mni', 'talairach', or 'native'
is_talairach	bool	Whether input was Talairach
tal_target	list[float] | None	Original Talairach coords (if applicable)
nii_path	str	Path to the NIfTI file used
atlas_shape	tuple	NIfTI volume shape
voxel_size_mm	tuple	Voxel dimensions in mm
threshold_used	float	Intensity threshold applied
surface_thickness_mm	float	Erosion depth used
to_dict()	dict	JSON-serializable representation

run_full() - single target, with volumetric arrays

Same signature as run(). Returns a PipelineState (extends ScalpResult) with additional NumPy arrays for visualization:

Attribute	Description
data	Raw NIfTI volume (float64)
affine	NIfTI affine matrix (4×4)
mask	Binary head mask (uint8)
filled	Hole-filled mask (uint8)
eroded	Eroded mask (uint8)
surface	Surface shell: filled XOR eroded (uint8)
scalp_surface_mni	(N, 3) float64 — all surface voxel centers in MNI mm

run_multi() / run_full_multi() - multiple targets

from brain2scalp.core import run_multi, run_full_multi

# Lightweight - one atlas load, one ScalpResult per target
results = run_multi(
    nii_path="head.nii",
    targets=[[-46.0, 20.0, 32.0], [10.0, -20.0, 45.0]],
)

# With arrays - returns list[PipelineState]
states = run_full_multi(
    nii_path="head.nii",
    targets=[[-46.0, 20.0, 32.0], [10.0, -20.0, 45.0]],
    is_talairach=False,
    surface_thickness_mm=1.5,
)

---

Output Formats

Output format is determined by the file extension passed to --output. Without --output, results are printed as text to stdout.

Text (.txt)

When Talairach input is used, tal_x/y/z columns are added automatically.

  target  mni_x   mni_y   mni_z   entry_x  entry_y  entry_z  dist_mm
       0  -46.0    20.0    32.0    -62.3     35.7     47.5     32.5
       1   10.0   -20.0    45.0     14.8    -28.1     58.3     19.2

JSON (.json)

A single result is returned as an object; multiple results as an array. When Talairach input is used, a tal_original field is added.

{
  "target": 0,
  "mni":  { "x": -46.0, "y": 20.0,  "z": 32.0 },
  "entry": { "x": -62.3, "y": 35.7,  "z": 47.5 },
  "distance_mm": 32.5
}

CSV (.csv)

Same columns as text. tal_x/y/z columns included when Talairach input was used.

target,mni_x,mni_y,mni_z,entry_x,entry_y,entry_z,dist_mm
0,-46.0,20.0,32.0,-62.3,35.7,47.5,32.5

Target file format (--target-file)

Plain CSV, one target per row. Header row is auto-detected and skipped.

x,y,z
-46,20,32
10,-20,45

---

Visualization

Requires pip install "brain2scalp[viz]".

Pipeline steps (--vizSteps)

6-panel figure showing each stage of the head-mask pipeline: raw data → binary mask → hole-filled → eroded → surface shell → overlay with target and entry point.

---

Orthogonal slice view (--vizTarget)

3-panel sagittal / coronal / axial slices centered on the brain target, with the projected scalp entry point overlaid.

---

Interactive 3D viewer (--viz3d)

Plotly HTML viewer with isosurface rendering of the head, scalp surface point cloud, and brain target / scalp entry markers. Includes threshold and opacity sliders. Supports up to 8 targets. Stride (settable via --stride) controls voxel subsampling for the isosurface: lower values give finer geometry but increase file size and browser render time; higher values are faster and lighter.

---

Package Layout

brain2scalp/
├── brain2scalp/
│   ├── core/
│   │   ├── models.py        ScalpResult, PipelineState
│   │   ├── transform.py     Talairach ↔ MNI (Lancaster 2007)
│   │   ├── atlas.py         Load, reorient, validate NIfTI
│   │   ├── mask.py          Threshold → fill → erode → surface
│   │   ├── projection.py    Ray-cast + nearest-neighbour fallback
│   │   └── pipeline.py      run() / run_full() / run_multi() / run_full_multi()
│   ├── cli/
│   │   ├── main.py          argparse entry point
│   │   └── formatters.py    text / JSON / CSV output
│   └── viz/
│       ├── slice2d_view.py  Pipeline steps + orthogonal slice figures
│       └── view_3d.py       Plotly interactive 3D viewer
├── tests/
│   ├── test_transform.py
│   ├── test_mask.py
│   ├── test_projection.py
│   ├── test_pipeline.py
│   ├── test_cli_helpers.py
│   └── test_formatters.py
└── assets/
    └── b2s_logo.png

---

[!WARNING] Research use only. This tool has not been validated for clinical use and must not be used to guide patient treatment, clinical diagnosis or medical decision-making. For any clinical application, always rely on a qualified clinician and a certified neuronavigation system.

---

License

MIT — see LICENSE.

Copyright 2026 Leonardo Politi.