Z-Rad

Extract quantitative features from medical images with a desktop interface or Python API.
CT, PET, MR, MG, US, and RTDOSE · DICOM and NIfTI · Windows, macOS, and Linux
Developed by the Department of Radiation Oncology at University Hospital Zurich

Full IBSI I preprocessing and feature coverage · All IBSI II filters

Download · Documentation · Python quickstart · Examples · Validation

From images to a feature table

Use preprocessing, filtering, and feature extraction independently or combine them into a complete workflow.

Supported images and masks

Use scalar image volumes with masks on the same physical voxel grid. See the data-format and folder-layout guide for input restrictions, case organization, and structure selection.

Get started

Desktop application

1. Open Z-Rad releases and choose a release.

2. Download the asset for your platform:

   Platform	Release asset	Launch
   Windows	z-rad-<release-tag>-windows.exe	Run the executable.
   Apple Silicon macOS	z-rad-<release-tag>-macos-arm64.zip	Extract the archive and open Z-Rad.app.

3. Follow the GUI quickstart to select your input data, configure processing, and run your first analysis.

The macOS app is currently unsigned and unnotarized, so Gatekeeper may show a warning. For Linux, Intel Macs, or the current source version, follow Run from source to install and launch the GUI.

Run the workflow interactively through the graphical user interface or automate it with the Python API.

Python quickstart

Requires Python 3.11 or newer. Install the published package:

python -m pip install z-rad

The example below requires your own NIfTI image and binary ROI mask on the same physical voxel grid. Replace the two paths with your files. It extracts intensity statistics without resampling or filtering:

from zrad.image import Image
from zrad.preprocessing import IntensityMaskBuilder, RoiData
from zrad.radiomics import Radiomics

image = Image.from_nifti("path/to/image.nii.gz")
mask = Image.from_nifti_mask("path/to/mask.nii.gz", reference=image)
roi = IntensityMaskBuilder().apply(RoiData(image=image, morphological_mask=mask))
features = Radiomics().extract_features(roi_data=roi, families=["intensity_statistics"])
print(features["stat_mean"])

The result is a dictionary of feature names and values. To try a complete example with supplied data and an expected result, follow the bundled phantom example.

In the Python single-ROI API, texture and intensity-volume histogram (IVH) features require additional preparation; see the full Python workflow.

IBSI validation

Automated benchmarks compare Z-Rad results against published reference data from the Image Biomarker Standardisation Initiative (IBSI):

• IBSI I: digital-phantom features and CT configurations A–E, including preprocessing diagnostics.
• IBSI II: digital-phantom filter response maps and features from filtered CT images.
• IBSI-SUV: SUV conversion checks using valid and intentionally invalid digital reference objects.

Implementation coverage describes available operations; benchmark agreement applies to the tested configurations and features. See IBSI coverage and limitations for the scope, comparison rules, and unavailable references, and execution reports and reproduction to inspect results for a particular revision.

Contribute and get in touch

Found a bug or have a feature request? Open an issue. To contribute code or documentation, start with CONTRIBUTING.md.

For questions or research collaborations, contact zrad@usz.ch.