Tutorial 10: Detecting Filamentous Fungi#

Filamentous fungi like Neurospora grow as branching networks of hyphae, not compact round colonies. Standard threshold detectors struggle with this morphology because hyphae are thin and have varying intensity. PhenoTypic includes a specialized pipeline for exactly this scenario.

What you will learn:

  1. Load a filamentous fungi plate image

  2. See why standard detectors miss thin hyphae

  3. Use FilamentousFungiPipeline for end-to-end processing

  4. Visualize fungal detection results

Load the Fungi Plate#

PhenoTypic ships a Neurospora filamentous fungi plate image. Let’s take a look at the spreading hyphal morphology.

[1]:

from phenotypic.data import load_fungi_plate

plate = load_fungi_plate()
plate.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

Notice the spreading mycelium thin, branching filaments radiating from each inoculation point. This is very different from the compact round yeast colonies we have worked with so far.

Protocol Prerequisite: Inoculum Density#

The detection algorithm anchors every colony from its inoculation centre (see How It Works). For this to succeed, the spore density of the inoculum must be high enough to produce a strong, well-defined signal that the blob detector and downstream branch calculations can work from. A faint or sparse deposit will not generate sufficient contrast for reliable detection, and if the inoculum step fails, all downstream assignment fails with it.

Ideally, the inoculum should be uniformly bright. Non-uniform deposits can sometimes be compensated by preprocessing steps for example, HomomorphicFilter for illumination gradients or the detector’s built-in GMM core extraction for fuzzy edges but severe patchiness may still cause detection failures or split a single inoculum into multiple detections.

Noise impacts branch detection much more than with yeast#

[2]:

from phenotypic.correction import StableDenoise

denoised = StableDenoise().apply(plate)
denoised.detect_mat.dash()

Data type cannot be displayed: application/vnd.plotly.v1+json

We’re looking for very fine structures in the image, 2 to 4 px wide branches. This makes detection very sensitive to noise.

Why Standard Detectors Struggle#

Let’s see what happens when we apply OtsuDetector to this plate.

[3]:

from phenotypic.detect import OtsuDetector
from phenotypic.correction import StableDenoise

test = OtsuDetector(ignore_borders=False).apply(plate)
test.gray.dash(show_grid=False)

Data type cannot be displayed: application/vnd.plotly.v1+json

The Otsu threshold fails to capture a lot of the structure due to the noise. Also

Note: Because this is a center crop of a larger plate, some of the branches touch the border, hence we set ignore_borders=False. Usually this is on by default.

Use FilamentousFungiPipeline#

FilamentousFungiPipeline is specifically designed for branching fungal morphology. It chains:

  1. BM3D denoising removes noise while preserving thin filaments

  2. Homomorphic filtering corrects uneven illumination

  3. FilamentousFungiDetector two-stage detection with Dijkstra reconnection to capture fragmented hyphal branches

[4]:

from phenotypic.prefab import FilamentousFungiPipeline
from phenotypic.detect import ManualGridDetector, FilamentousFungiDetector

plate = load_fungi_plate()

fungi_pipeline = FilamentousFungiPipeline(
        inoculum_detector=ManualGridDetector(coord1=(230, 240),
                                             coord2=(630, 640),
                                             shape="disk",
                                             width=100),
        ignore_borders=False
)
plate = fungi_pipeline.apply(plate)

fig = plate.detect_mat.dash(overlay=True)
fig.show()

Data type cannot be displayed: application/vnd.plotly.v1+json

Much better! The specialized pipeline captures the full extent of the mycelium, including thin outer branches that the Otsu detector missed. Most importantly, it reconnects those missed fragments from before

[5]:

print(f"Detected fungal colonies: {plate.num_objects}")

Detected fungal colonies: 8

Using radial expansion as a metric of growth#

[6]:

from phenotypic.measure import MeasureSymmetricZones

symmetric_radius = MeasureSymmetricZones()
meas = symmetric_radius.measure(plate)

[7]:

column = symmetric_radius.inspect()
column

[7]:

A Note on FrangiVesselness#

If you want to build a custom pipeline for filamentous organisms, the FrangiVesselness enhancer is a useful building block. It enhances tubular structures (hyphae, branches) in detect_mat using Hessian-based vesselness filtering.

from phenotypic.enhance import FrangiVesselness

frangi = FrangiVesselness(sigmas=(0.5, 1.0, 1.5), black_ridges=False)

The FilamentousFungiPipeline uses a similar approach internally, combined with phase congruency and minimum-cost path reconnection.

Summary#

You have detected filamentous fungi using PhenoTypic’s specialized pipeline:

  • Standard detectors (like OtsuDetector) miss thin hyphae because a single threshold cannot capture varying filament intensities.

  • ``FilamentousFungiPipeline`` handles this with BM3D denoising, homomorphic filtering, and a two-stage detector with Dijkstra reconnection.

  • ``FrangiVesselness`` is available for custom pipeline building.

Congratulations you have completed the tutorial series! You now know how to:

  1. Load and inspect plate images

  2. Detect colonies with thresholding and specialized detectors

  3. Enhance images before detection

  4. Build reusable pipelines

  5. Work with grid plates

  6. Batch-process many plates from the CLI

  7. Measure and export colony features

  8. Use prefab pipelines for common scenarios

  9. Assess image quality

  10. Handle filamentous fungi

For task-specific recipes, head to the How-To Guides. For deeper understanding, see the Explanation pages.

[ ]: