Pyimagej Fiji Bridge

Prerequisites

• Python packages: pyimagej, scyjava, numpy, pandas
• Java: Java 8 or Java 11 (Java 17 is not supported); use conda for reliable Java management
• Fiji/ImageJ2: Downloaded automatically on first init, or specify a local Fiji installation path
• Environment: conda environment strongly recommended; pip-only installs often have JVM path issues

# Recommended: conda installation
conda create -n pyimagej -c conda-forge pyimagej openjdk=11
conda activate pyimagej

# Install additional dependencies
pip install pandas tifffile

# Verify
python -c "import imagej; ij = imagej.init('sc.fiji:fiji', mode='headless'); print(ij.getVersion())"

Quick Start

import imagej
import numpy as np

# Initialize Fiji in headless mode (downloads on first run, ~500 MB)
ij = imagej.init("sc.fiji:fiji", mode="headless")
print(f"ImageJ version: {ij.getVersion()}")

# Create a test image, process with Gaussian blur via Ops, convert back
arr = np.random.randint(0, 1000, (256, 256), dtype=np.uint16)
imp = ij.py.to_imageplus(arr)
blurred = ij.op().filter().gauss(imp.getProcessor(), 2.0)
result = ij.py.from_imageplus(imp)
print(f"Processed array shape: {result.shape}, dtype: {result.dtype}")

Core API

Module 1: Initialization

PyImageJ must be initialized once per Python session. The mode and endpoint determine which ImageJ distribution and GUI behavior to use.

import imagej

# Headless Fiji — most common for scripts and batch jobs
ij = imagej.init("sc.fiji:fiji", mode="headless")

# GUI mode — opens the Fiji window (requires a display)
ij = imagej.init("sc.fiji:fiji", mode="gui")

# Local Fiji installation — faster startup, no download
ij = imagej.init("/path/to/Fiji.app", mode="headless")

# Specific Fiji version
ij = imagej.init("sc.fiji:fiji:2.14.0", mode="headless")

# Bare ImageJ2 without Fiji plugins
ij = imagej.init("net.imagej:imagej", mode="headless")

print(f"ImageJ version: {ij.getVersion()}")
print(f"Headless: {ij.ui().isHeadless()}")

Module 2: Image I/O

Open and save images using ImageJ's I/O layer (which includes Bio-Formats for proprietary formats) and convert between ImageJ and NumPy representations.

import imagej
import numpy as np

ij = imagej.init("sc.fiji:fiji", mode="headless")

# Open any format Bio-Formats supports: CZI, LIF, ND2, ICS, TIFF, etc.
imp = ij.io().open("/data/experiment.czi")
print(f"Dimensions: {imp.getDimensions()}")   # [W, H, C, Z, T]
print(f"nSlices: {imp.getNSlices()}, nFrames: {imp.getNFrames()}")

# Save image
ij.io().save(imp, "/data/output.tif")
print("Saved output.tif")

# NumPy ↔ ImageJ conversion
arr = np.zeros((100, 100), dtype=np.uint16)
arr[30:70, 30:70] = 1000   # bright square

# NumPy → ImagePlus
imp = ij.py.to_imageplus(arr)
print(f"ImagePlus: {imp.getWidth()}×{imp.getHeight()}, type={imp.getType()}")

# ImagePlus → NumPy (returns a view where possible)
arr_back = ij.py.from_imageplus(imp)
print(f"NumPy array: shape={arr_back.shape}, dtype={arr_back.dtype}")

# Multi-channel array: shape (C, H, W)
rgb = np.random.randint(0, 255, (3, 256, 256), dtype=np.uint8)
imp_rgb = ij.py.to_imageplus(rgb)
print(f"Channels: {imp_rgb.getNChannels()}")

Module 3: Macro Execution

Run ImageJ macro language (IJM) snippets or macro files. Macros execute inside the ImageJ environment and can call any built-in ImageJ command.

import imagej

ij = imagej.init("sc.fiji:fiji", mode="headless")

# Run an inline macro string
ij.macro.run("print('Hello from ImageJ macro');")

# Run a macro with options string (key=value pairs)
# Options string mirrors the dialog parameters of ImageJ commands
macro_code = """
run("Gaussian Blur...", "sigma=2");
run("Auto Threshold", "method=Otsu white");
"""
ij.macro.run(macro_code)

# Run a macro file from disk
ij.macro.runMacroFile("/scripts/my_analysis.ijm")

# Run macro that returns a value via getResult or output string
result = ij.macro.run("""
x = 42 * 2;
return x;
""")
print(f"Macro returned: {result}")

# Macro with current image: open → process → measure
ij.io().open("/data/cells.tif")   # sets current active image

measure_macro = """
run("Set Measurements...", "area mean min integrated redirect=None decimal=3");
run("Analyze Particles...", "size=50-Infinity display clear summarize");
"""
ij.macro.run(measure_macro)
print("Analyze Particles complete; results in Results table")

Module 4: ImageJ Ops

ImageJ Ops is a framework of 150+ image processing operations with type-safe dispatch. Ops work on ImgLib2 Img objects and are the preferred way to call image processing algorithms programmatically.

import imagej
import numpy as np

ij = imagej.init("sc.fiji:fiji", mode="headless")

arr = np.random.randint(100, 900, (512, 512), dtype=np.uint16)
img = ij.py.to_java(arr)   # converts to ImgLib2 RandomAccessibleInterval

# Gaussian blur
blurred = ij.op().filter().gauss(img, 2.0)
blurred_np = ij.py.from_java(blurred)
print(f"Blurred: {blurred_np.shape}")

# Otsu threshold → binary image
binary = ij.op().threshold().otsu(img)
binary_np = ij.py.from_java(binary)
print(f"Binary unique values: {np.unique(binary_np)}")

# Morphological operations
from jnius import autoclass
BitType = autoclass("net.imglib2.type.logic.BitType")
opened = ij.op().morphology().open(binary, [3, 3])
opened_np = ij.py.from_java(opened)
print(f"After opening: {opened_np.shape}")

# Statistics ops
mean_val = ij.op().stats().mean(img)
std_val  = ij.op().stats().stdDev(img)
print(f"Mean intensity: {mean_val:.1f}, StdDev: {std_val:.1f}")

# Math ops: multiply image by scalar
scaled = ij.op().math().multiply(img, ij.py.to_java(2.0))
print(f"Scaled max: {ij.py.from_java(scaled).max()}")

Module 5: Plugin and Command Calls

SciJava commands are the primary way to invoke Fiji plugins programmatically. Commands accept a dict of named parameters mirroring the plugin dialog.

import imagej

ij = imagej.init("sc.fiji:fiji", mode="headless")

# Open a file using Bio-Formats opener command
future = ij.command().run(
    "loci.plugins.LociImporter",
    True,
    {"id": "/data/image.lif", "open_files": True, "autoscale": True}
)
module = future.get()
imp = module.getOutput("imp")
print(f"Opened via Bio-Formats: {imp.getDimensions()}")

# Run Analyze Particles as a SciJava command
ij.io().open("/data/binary_mask.tif")

future = ij.command().run(
    "ij.plugin.filter.ParticleAnalyzer",
    True,
    {
        "minSize": 50.0,
        "maxSize": float("inf"),
        "options": 0,      # SHOW_NONE
        "measurements": 1,  # AREA
    }
)
future.get()
print("Analyze Particles command complete")

# Alternatively, run via macro string for simpler plugin invocation
ij.macro.run("""
run("Analyze Particles...", "size=50-Infinity display clear summarize");
""")

Module 6: Results Table and ROI Analysis

Retrieve measurement results from ImageJ's Results table and ROI Manager after running Analyze Particles or other measurement commands.

import imagej
import pandas as pd

ij = imagej.init("sc.fiji:fiji", mode="headless")

# After running Analyze Particles, read the Results table
def results_to_dataframe(ij) -> pd.DataFrame:
    """Convert ImageJ Results table to pandas DataFrame."""
    rt = ij.ResultsTable.getResultsTable()
    if rt is None or rt.size() == 0:
        return pd.DataFrame()
    headings = list(rt.getHeadings())
    data = {col: [rt.getValue(col, i) for i in range(rt.size())]
            for col in headings}
    return pd.DataFrame(data)

# Run segmentation + measurement macro
ij.io().open("/data/cells.tif")
ij.macro.run("""
run("Gaussian Blur...", "sigma=1.5");
setAutoThreshold("Otsu dark");
run("Convert to Mask");
run("Analyze Particles...", "size=20-Infinity display clear");
""")

df = results_to_dataframe(ij)
print(f"Found {len(df)} objects")
print(df[["Area", "Mean", "IntDen"]].describe())
df.to_csv("particle_measurements.csv", index=False)
print("Saved particle_measurements.csv")

# Access the ROI Manager
def get_roi_manager(ij):
    """Return the ImageJ ROI Manager instance, creating if needed."""
    RoiManager = ij.py.jclass("ij.plugin.frame.RoiManager")
    rm = RoiManager.getInstance()
    if rm is None:
        rm = RoiManager(False)   # headless=False means no GUI window
    return rm

rm = get_roi_manager(ij)
roi_count = rm.getCount()
print(f"ROIs in manager: {roi_count}")

# Extract bounding boxes for all ROIs
rois = []
for i in range(roi_count):
    roi = rm.getRoi(i)
    bounds = roi.getBounds()
    rois.append({"index": i, "x": bounds.x, "y": bounds.y,
                 "width": bounds.width, "height": bounds.height})
roi_df = pd.DataFrame(rois)
print(roi_df.head())

Common Workflows

Workflow 1: Automated Fluorescence Quantification

Goal: Open a multi-channel TIFF stack, apply Gaussian blur, threshold nuclei channel, run Analyze Particles, and export per-cell measurements as CSV.

import imagej
import pandas as pd
import numpy as np
from pathlib import Path

ij = imagej.init("sc.fiji:fiji", mode="headless")

def quantify_nuclei(tiff_path: str, output_csv: str,
                    channel: int = 1, sigma: float = 1.5,
                    min_size: int = 50) -> pd.DataFrame:
    """
    Segment and measure nuclei in a fluorescence TIFF.

    Parameters
    ----------
    tiff_path  : path to single- or multi-channel TIFF
    output_csv : where to save results
    channel    : 1-based channel index for nuclear stain (e.g., DAPI)
    sigma      : Gaussian blur radius in pixels
    min_size   : minimum nucleus area in pixels
    """
    # Step 1: Open image
    imp = ij.io().open(tiff_path)
    print(f"Loaded: {Path(tiff_path).name}  dims={imp.getDimensions()}")

    # Step 2: Extract channel if multi-channel
    if imp.getNChannels() > 1:
        imp.setC(channel)

    # Step 3: Apply Gaussian blur and threshold via macro
    ij.macro.run(f"""
selectWindow("{imp.getTitle()}");
run("Gaussian Blur...", "sigma={sigma}");
setAutoThreshold("Otsu dark");
run("Convert to Mask");
run("Fill Holes");
run("Watershed");
""")

    # Step 4: Measure
    ij.macro.run(f"""
run("Set Measurements...", "area mean min centroid integrated shape redirect=None decimal=3");
run("Analyze Particles...", "size={min_size}-Infinity display clear include summarize");
""")

    # Step 5: Collect results
    rt = ij.ResultsTable.getResultsTable()
    if rt is None or rt.size() == 0:
        print("No objects detected")
        return pd.DataFrame()

    headings = list(rt.getHeadings())
    df = pd.DataFrame(
        {col: [rt.getValue(col, i) for i in range(rt.size())]
         for col in headings}
    )
    df["source_file"] = Path(tiff_path).stem
    df.to_csv(output_csv, index=False)
    print(f"Saved {len(df)} measurements → {output_csv}")
    return df


df = quantify_nuclei(
    tiff_path="/data/experiment_dapi.tif",
    output_csv="nuclei_measurements.csv",
    channel=1,
    sigma=1.5,
    min_size=50
)
print(df[["Area", "Mean", "Circ."]].describe())

Workflow 2: Batch Fiji Macro Processing

Goal: Process a folder of images with an existing Fiji .ijm macro file, collect the Results table from each image into a single DataFrame.

import imagej
import pandas as pd
from pathlib import Path

ij = imagej.init("sc.fiji:fiji", mode="headless")

def run_macro_on_image(ij, image_path: str, macro_file: str) -> pd.DataFrame:
    """Open one image, run a macro file, return its Results table."""
    ij.io().open(image_path)

    # Clear any previous results before running
    ij.macro.run("run(\"Clear Results\");")

    # Run macro file (macro must operate on the active image)
    ij.macro.runMacroFile(macro_file)

    rt = ij.ResultsTable.getResultsTable()
    if rt is None or rt.size() == 0:
        return pd.DataFrame()
    headings = list(rt.getHeadings())
    return pd.DataFrame(
        {col: [rt.getValue(col, i) for i in range(rt.size())]
         for col in headings}
    )


input_dir  = Path("/data/images")
macro_file = "/scripts/measure_cells.ijm"
output_csv = "batch_results.csv"

all_results = []
image_files = sorted(input_dir.glob("*.tif"))

for img_path in image_files:
    print(f"Processing: {img_path.name}")
    df = run_macro_on_image(ij, str(img_path), macro_file)
    if not df.empty:
        df["filename"] = img_path.name
        all_results.append(df)

    # Close all windows to free memory between images
    ij.macro.run("close('*');")

if all_results:
    combined = pd.concat(all_results, ignore_index=True)
    combined.to_csv(output_csv, index=False)
    print(f"Batch complete: {len(image_files)} images, "
          f"{len(combined)} total measurements → {output_csv}")