PyNeuronJ

English | 简体中文

What is this?

A Python implementation of NeuronJ for tracing nerve fibers in 2D grayscale images. Give it a start and an end point; it returns the centerline and its length. NDF files can be read, edited and saved for use in NeuronJ.

The algorithm follows Meijering et al. (2004). The default practical mode includes numerical corrections. Use mode="paper" for the paper formulas.

Installation

Python 3.10 or newer:

python -m pip install "pyneuronj[fast,examples]"

Requires NumPy and SciPy. The command also installs Numba for acceleration and Pillow for the example. For the library alone, use python -m pip install pyneuronj. To install from source, run python -m pip install -e ".[fast,examples]" in the repository.

Quick start

Results on the bundled image, using the same endpoints:

PyNeuronJ	NeuronJ 1.4.3

Red: centerline. Green: start. Blue: end.

from importlib.resources import files

import numpy as np
from PIL import Image
from pyneuronj import PyNeuronJ, load_ndf, save_ndf

with files("pyneuronj").joinpath("data/test.jpg").open("rb") as image_file:
    with Image.open(image_file) as source:
        image = np.asarray(source.convert("L"))

tracer = PyNeuronJ(image)
result = tracer.trace(start=(104, 156), end=(149, 316))

print(result.points_xy)       # Centerline coordinates
print(result.length_pixels)   # Length in pixels
save_ndf("test.ndf", result)

document = load_ndf("test.ndf")
document.tracings[0].label = "Fiber 1"
save_ndf("edited.ndf", document)

Coordinates are (x, y) pixel centers. For calibrated lengths, use result.length(pixel_size=(sx, sy)). Replace the image and endpoints to trace your own data.

NDF supports versions 1.0.0–1.4.3 and saves as 1.4.3. Types, colors, clusters, labels and segment boundaries are preserved. To load the saved annotation in NeuronJ, open the same image first.

The complete example can also be run from the command line:

python examples/basic.py

Validation

Compared with NeuronJ 1.4.3 on 196 images and 9,305 segments, using the same endpoints.