RingSentry

Reproducible preprocessing and quality control for 2D diffraction detector images

SAXS · WAXS · SXRD · GIWAXS

RingSentry is a local desktop application and Python processing core for inspectable preprocessing of two-dimensional diffraction detector images before downstream integration, fitting, texture analysis, or other quantitative interpretation. It combines multi-format input/output, non-mutating quality-control suggestions, an explicitly ordered numerical pipeline, previews, batch reports, a conservative CBF zero-value repair workflow, and detector-geometry/Q conversion tools.

二维衍射图预处理：处理顺序、风险提示和参数进入日志与报告，避免静默改变数据含义。

Why RingSentry

Detector data often arrive in instrument-specific formats and must be checked, corrected, cropped, masked, transformed, and exported before analysis. RingSentry makes that preparation path visible in one interface and keeps the scientific data path separate from display-only exports.

RingSentry does not perform azimuthal integration, peak fitting, structure refinement, or automatic physical interpretation. Its QC rules report transparent numerical evidence and suggestions; they do not silently change processing parameters.

Install

Use a virtual environment and install the project itself, not only its dependency list.

Windows PowerShell

py -3.12 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install .
ringsentry

The bundled START_RingSentry.cmd, START_RingSentry.bat, and 双击启动_RingSentry.cmd launchers are also available for Windows.

macOS or Linux

python3 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install .
ringsentry

The GUI requires Tk. Some Linux distributions package this separately (for example, python3-tk). The current maintainer workflow is primarily validated on Windows; use the headless example below to verify the numerical core on another platform.

For development or review:

python -m pip install -e ".[test]"
python -m ruff check .
python -m pytest -q
python -m build

Quick start

1. Select an input directory recursively or choose individual files.
2. Select an output directory; the GUI otherwise proposes _converted.
3. Choose 统计文件 and run 自动质控 / 分析样本.
4. Use the coordinate preview to select an ROI, then use 单图模式 (Single image) to inspect the same complete numerical pipeline that batch processing applies. When PNG output is selected, the single-image view also applies its display rendering options.
5. Select quantitative matrix outputs (EDF or NPY are the safest defaults) or display/text outputs as needed.
6. Choose 开始转换, review the preflight messages, and inspect the generated run_report_*.txt.

Preflight quality control samples at most five input files. During conversion, the worker emits per-file QC and writer messages for files it can load; the GUI copies those messages and the selected processing/output settings into the run report. The report is provenance for the run, not proof that every warning has been resolved.

The numerical order is fixed and documented: dark subtraction → flat correction → background offset → ROI → mask → absolute and percentile clipping → negative clipping → hot-pixel suppression → rotation/flips → binning → intensity transform → gamma → normalization.

Reproducible headless example

This example generates its own deterministic synthetic ring image. It is not an experimental result or a performance benchmark.

python examples/minimal_preprocessing.py --output-dir example_output

Expected checks include a 128 × 128 finite input, a 64 × 64 finite processed array after 2× binning, and these outputs:

• synthetic_raw.npy and synthetic_processed.npy: quantitative arrays;
• synthetic_processed_display.png: display-only rendering;
• summary.json: seed, processing options, QC facts, shapes, ranges, and output provenance.

Supported data

Implemented input path	Reader and current validation scope
TIFF (.tif, .tiff)	tifffile/imageio; repository fixtures and round-trip tests exercise TIFF files
MCCD, MARCCD	Routed through the TIFF-family reader by header or suffix; no dedicated MCCD/MARCCD fixture is included
HDF5 and NeXus	h5py with a configurable dataset path; callers must select a single 2D dataset
EDF	Strict uncompressed 2D reader; project and FabIO interoperability tests exercise supported files
CBF	FabIO, exercised by repository round-trip and exceptional-value tests
ADSC/Bruker IMG, MAR3450, SFRM	FabIO reader routes are implemented; detector-specific sample compatibility is not yet covered by repository fixtures

Output	Intended use
EDF, NPY	Quantitative matrix preservation
TIFF	Scientific-image interoperability; CBF integer dtype conversion is logged
CSV/DAT matrix or x-y-intensity columns	Text-based interchange; retain the GUI run report for XY option provenance
PNG	Display and rapid inspection only, never the quantitative matrix

Safety boundaries

• CBF overexposure repair can replace stored zero values with a configured saturation-like value only after the user confirms the acquisition convention. Zero may instead represent a beamstop, module gap, mask, or genuine low count. A formal GUI repair requires a rule-evidence note, writes a separate copy, and requires read-back verification; the GUI rejects original-file overwrite. The lower-level API also rejects original overwrite; disabled verification remains an explicit escape hatch, and output written without verification is marked repaired_unverified. SHA-256 fields are populated only when compute_sha256 is enabled. The tool cannot recover lost intensity.
• Validation indexes record SHA-256-linked original/output pairs as advisory evidence, but they never authorize automatic removal. RingSentry reports cleanup candidates and leaves archival or deletion to a separate, user-controlled process after independent review.
• Duplicate-name CBF candidates are likewise reported by name, size, and SHA-256 evidence; RingSentry does not move them automatically.
• Flat correction expects a relative detector-response map. RingSentry does not automatically normalize raw flat counts, so users requiring scale-preserving correction must normalize that map before processing.
• A mask may set invalid pixels to NaN. Floating-point NPY, EDF, and TIFF preserve IEEE NaN/Inf; CSV/DAT matrix exports replace them with zero and log separate counts. Third-party TIFF viewers may display non-finite pixels differently.
• Binning uses block means and crops non-divisible right/bottom edges.
• Percentile clipping, intensity transforms, gamma correction, and normalization alter numerical meaning and should be enabled only when scientifically justified.
• PNG is an 8-bit RGB view; use EDF or NPY for quantitative downstream work.

Documentation

• User guide: installation, complete GUI workflow, formats, processing semantics, CBF repair, Q tools, and troubleshooting.
• Core API: programmatic loading, QC, processing, writing, and diffraction-geometry interfaces.
• Contributing, Code of Conduct, and security policy.
• JOSS paper source.

Version and citation

The source tree identifies itself as 7.0.0. The latest public Git tag, GitHub Release, and version-specific Zenodo archive are currently v6.0.1; the concept DOI 10.5281/zenodo.19602728 resolves to that latest archived version until a new release is published. Do not describe v7.0.0 as an archived release until its tag, release, and Zenodo record exist.

Use CITATION.cff for the maintained author and software metadata. For reproducibility, cite the exact version-specific archive DOI once the version used in the analysis has been released.

Support and governance

Use GitHub Issues for reproducible bugs, focused feature proposals, and general usage questions. Follow SECURITY.md for vulnerabilities and avoid uploading private beamline data. RingSentry is currently maintained as a single-maintainer research project; decisions prioritize numerical transparency, regression tests, documented data semantics, and backward compatibility. Maintenance and support are provided on a best-effort basis.

License

RingSentry is released under the MIT License.