Python tools for standard Winston–Lutz analysis and CT-referenced, off-isocentre MultiWLT. The desktop application uses pylinac; the new multiwlt/ research module explicitly subtracts the expected CT/plan offset before evaluating measured deviations.
Windows without Python: download the portable MultiWLT GUI. Extract the complete ZIP, start MultiWLT.exe, select MV images, matching reference and delivery records, then create the PDF/CSV/JSON report. Quick start and initial-data reproduction. This dedicated GUI uses the CT-corrected method; the older general application's Multi-Ball tab remains the legacy raw-distance method.
| Workflow | Entry point | What it reports |
|---|---|---|
| Standard single-ball WLT | Desktop GUI, Standard WLT tab | Standard pylinac results and PDF report |
| Existing multi-ball exploration | Desktop GUI, Multi-Ball Off-Iso tab | Detected ball-to-field distances and aperture previews |
| CT-referenced MultiWLT | multiwlt/analyse.py |
Measured minus expected vectors, local 3D ray envelopes, compact PDF |
The existing GUI multi-ball tab does not apply the new CT-reference correction. Use the new module's CLI for the corrected workflow. The original desktop application remains available; its German guide is preserved.
Four acquired MVHighQuality images, three balls, twelve ball–field pairs. The initial test was performed on 25 September 2026 by Dr. Sebastian Schäfer in Leipzig. The largest additional 2D displacement was 0.508 mm; one field-centre method comparison remains flagged for review. These are initial research results, not clinical acceptance criteria.
Start with the English MultiWLT guide. The public MultiWLT module also generates English figures and reports. It includes the measured results, expected-versus-observed plots, a synthetic demo, a machine-name adapter and a tool for inspecting your own CT and plan.
Expected offsets are specific to the CT ball centres, plan isocentre, angles and apertures. Do not reuse our offsets for a different CT/plan combination. Even with the same phantom model, verify your own ball layout and derive your own reference. inspect_phantom.py performs that check; unsupported ball arrangements, openings or image geometry require adapting and validating the logic.
git clone https://github.com/Kiragroh/Python_WinstonLutzAnalyzer.git
cd Python_WinstonLutzAnalyzer/multiwlt
python -m pip install -r requirements.txt
python demo.py --output demo_output
python -m pytest tests -qThe demo uses labelled synthetic images and an incomplete synthetic geometry fixture, not a deliverable treatment plan. The acquired phantom dataset is now available as a GitHub Release (approximately 84 MB): CT, source RTPLAN, RTSTRUCT, four MV images and four delivery records. Download checksums and file counts are in data/manifest.json. Large DICOM files are stored as release assets; institution-specific deployment configuration is not included.
We warmly thank Paul Rétif for his ideas, his earlier ImageJ script and his 2024 paper on cine EPID and off-axis Winston–Lutz testing, which helped inspire this project. See the full acknowledgement and methodological distinction.
Python 3.13 is the tested interpreter. From the repository root:
py -3.13 -m pip install -r requirements.txt
.\start_open_gui.cmdThe GUI supports folder selection, standard WLT reports, exploratory multi-ball aperture views and local history. Output and saved settings remain local in output/. See the desktop guide for startup checks, CLI options and screenshots.
Only the initial four-view MultiWLT measurements are presented as acquired results. Couch/collimator variations, comparisons with and without ExacTrac correction, conventional WLT on BB01, and DCA/cine are research outlook. This repository does not operate a treatment machine or establish a clinical tolerance.
Reproducible, deidentified cases, known-shift checks and independent localisation comparisons are welcome. Document the CT/plan reference and acquisition geometry with each case. The software is released under the MIT licence; dependency and media details for the extension are in THIRD_PARTY.md.
MultiWLT downloads are separated into stable CT/RTSTRUCT, versioned plan/reference and initial measurements; plan revisions do not require uploading the CT again.