Fibre Explorer PREVIEW

2D fibre and matrix analysis

Analysis stays on your computer

FIBRE ANALYSIS

Single-image analysis

⇄ Linked views
Load a TIFF or try the synthetic example. Nothing is uploaded.
A

Source image

INPUT
No image loadedOpen a TIFF image to begin.
Original pixel intensities
B

Fibre overlay

TRACES
Fibre traces will appear here
Traces on the analysis image
C

Local trace density

DENSITY
Run the analysis to calculate local density
Overlapping component neighbourhoods
D

Local orientation

DIRECTION
Run the analysis to calculate orientation
Legacy arithmetic mean · degrees
E

Distance to Canny edges

DISTANCE
Run the analysis to calculate edge distances
Euclidean distance · pixels
F
Switch between additional analysis maps
Connected regions above intensity threshold
Scroll to zoom · drag to pan · all views move togetherNo analysis yet

Measurements

No results
Trace components—
Trace area fraction—
Mean major axis—
Gap area fraction—

Trace components can be fragments of a fibre. Major axis is not path length; gaps are measured from Canny edges.

What it means

How it affects the analysis

For your matrix images

Help, methods and references

Fibre Explorer adapts the Trace Ridges approach for browser-local 2D analysis. Selected images are processed on your computer and are not uploaded.

References

  1. Trace Ridges. Arafat Y, Cuesta-Apausa C, Castellano E, Reyes-Aldasoro CC (2025). Fibre tracing in biomedical images: An objective comparison between seven algorithms. PLOS ONE. DOI: 10.1371/journal.pone.0320006.
  2. Spatial smoothing / Getis–Ord. Gerber R, Griner J, Incicau D, Guglietta S, Krieg C, Robinson MD (2025). Spatially-smoothed quantification improves cell typing in imaging mass cytometry datasets. bioRxiv preprint, version 1; not peer reviewed in the linked PMC record. DOI: 10.64898/2025.12.16.693927.

Adaptation of Trace Ridges

Trace Ridges combines watershed ridges, Canny edges and morphological filtering to identify fibre-like traces. This app uses the supplied Python implementation, running through Pyodide, rather than the original MATLAB implementation.

The copied Python tracing code is unchanged. Browser output has been checked against that Python implementation on example images; that check establishes port consistency, not biological accuracy or equivalence to the original MATLAB results.

Is Getis–Ord smoothing or denoising?

The Gerber et al. preprint describes GZ as neighbourhood smoothing plus intensity normalisation. It reports context-dependent effects on IMC cell typing: smoothing can help, but it can also mix nearby signals. This supports a smoothing rationale; it does not establish accuracy for matrix-fibre tracing.

Here, Getis–Ord is an optional smoothing and local-contrast preprocessing step. It may reduce pixel-scale noise, but it can blur thin fibres or merge adjacent structures. Inspect the traces against the source image and compare with Raw intensity on representative regions.

How this implementation differs from the paper’s GZ

This app preserves the supplied notebook’s cautils G_classical behaviour: a uniform 3 × 3 kernel including the centre, replicated image boundaries, and integer rounding of local means for integer inputs. Its normalisation gives approximately (local 3 × 3 mean − image mean) / image standard deviation, with rounding where applicable.

The paper describes an eight-neighbour queen neighbourhood and a weighted Getis–Ord variance term involving squared weights. Our notebook-compatible scaling is not an exact implementation of that published normalisation. We additionally set negative values to zero and cap values at the chosen percentile (97 by default). Statistics are recomputed within each image or crop.

Outputs are transformed intensities, not p-values or calibrated hotspot significance scores. The app does not perform permutation testing, cell segmentation, marker aggregation or the study’s cell-typing evaluation.

Using the app

Open a TIFF, choose its plane and channel if necessary, and optionally draw a crop. Set the preprocessing and tracing parameters, then run. The ? buttons explain individual controls. Parameter changes leave the completed result visible until you rerun; exports retain that completed run’s settings.

In Batch folder mode, TIFFs in the selected folder and subfolders are analysed sequentially with the same captured settings. Whole images are used, without the single-image crop. Failed files are logged and processing continues. Stop after current image produces a partial ZIP; the full ZIP includes combined tables and per-image results.

Interpreting the measurements

Limits and reproducibility

TIFFs up to 16 megapixels can be opened; each analysis is limited to 4,194,304 pixels and at least 16 × 16 pixels. Calibration is entered manually and assumes square pixels. Tracing parameters stay in pixels. Crop boundaries and image-wide thresholds can change the results.

Keep the tab open and the computer awake during analysis. Download results before replacing them. Archives contain references, settings, software versions, the input-region checksum, CSV measurements, figures and numerical maps. Source-display contrast and the interface colour scheme do not change the analysis.

Export this analysis

Summary and per-component CSVs, run settings, numerical maps (.npy), and a six-panel PNG. Includes the analysed source crop for reproducibility.