Tutorial 8: Using Prefab Pipelines

Building a pipeline from scratch gives you full control, but PhenoTypic also ships prefab pipelines — pre-configured ImagePipeline subclasses tuned for common organisms and plate types. In this tutorial you will survey the available prefabs, apply one, and compare results.

What you will learn:

1. What prefab pipelines are and when to use them

2. Survey the available prefabs

3. Apply a prefab pipeline

4. Compare results from different prefabs

Imports

from phenotypic.data import load_synth_yeast_plate

Available Prefab Pipelines

Each prefab is an ImagePipeline subclass with operations and measurements already configured. Choose based on your organism and plate conditions.

Prefab	Best for	Strategy
HeavyOtsuPipeline	General-purpose yeast, clean plates	Multi-stage Otsu with refinement
HeavyWatershedPipeline	Touching or clustered colonies	Watershed segmentation
RoundPeaksPipeline	Round colonies, lightweight	Peak detection
HeavyRoundPeaksPipeline	Round colonies needing refinement	Extended peak detection
FilamentousFungiPipeline	Filamentous fungi (Neurospora, etc.)	BM3D denoise + specialized detector
GridSectionPipeline	Pre-tiled grid sections	Section-level processing

Apply HeavyOtsuPipeline

Let’s start with the most general-purpose option — HeavyOtsuPipeline. It chains Gaussian blur, CLAHE, median filtering, Sobel edge enhancement, Otsu detection, and several refinement steps (morphological opening, border removal, small-object removal, mask fill).

from phenotypic.prefab import HeavyOtsuPipeline

plate = load_synth_yeast_plate()
heavy_otsu = HeavyOtsuPipeline()
result_otsu = heavy_otsu.apply(plate)
result_otsu.dash(overlay=True)

PNG file does not have exif data.

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

print(f"HeavyOtsuPipeline detected {result_otsu.num_objects} colonies")

HeavyOtsuPipeline detected 95 colonies

Apply RoundPeaksPipeline

Now let’s try RoundPeaksPipeline — a lighter approach that uses peak detection to find circular colonies. It is faster but may miss irregular or faint colonies.

from phenotypic.prefab import RoundPeaksPipeline

plate2 = load_synth_yeast_plate()
round_peaks = RoundPeaksPipeline()
result_rp = round_peaks.apply(plate2)
result_rp.dash(overlay=True)

PNG file does not have exif data.

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

print(f"RoundPeaksPipeline detected {result_rp.num_objects} colonies")

RoundPeaksPipeline detected 96 colonies

Compare

Different prefabs produce different results on the same plate. The best choice depends on your colonies, your imaging conditions, and what you need to measure. A quick visual comparison and colony count helps you decide.

print(f"HeavyOtsuPipeline:   {result_otsu.num_objects} colonies")
print(f"RoundPeaksPipeline:  {result_rp.num_objects} colonies")

HeavyOtsuPipeline:   95 colonies
RoundPeaksPipeline:  96 colonies

Prefabs Include Measurements

Prefab pipelines come with measurements pre-configured, so you can call .apply_and_measure() directly — no need to add your own meas list.

plate3 = load_synth_yeast_plate()
df = heavy_otsu.apply_and_measure(plate3)
print(f"{len(df)} colonies measured across {df.shape[1]} features")
df.head()

PNG file does not have exif data.

95 colonies measured across 179 features

	Metadata_FileSuffix	Metadata_BitDepth	Metadata_ImageType	Metadata_ImageName	ObjectLabel	Bbox_CenterRR	Bbox_CenterCC	Bbox_MinRR	Bbox_MinCC	Bbox_MaxRR	...	Intensity_MaximumIntensity	Intensity_MeanIntensity	Intensity_MedianIntensity	Intensity_StandardDeviationIntensity	Intensity_CoefficientVarianceIntensity	Intensity_LowerQuartileIntensity	Intensity_UpperQuartileIntensity	Intensity_InterquartileRangeIntensity	Intensity_Density	Intensity_ConvexDensity
0	.png	8	GridImage	Synthetic96PlateWithObjects	1	66.379006	675.189503	38	652	96	...	0.943073	0.775636	0.911701	0.169254	0.218314	0.557704	0.923465	0.365762	0.775636	10.005758
1	.png	8	GridImage	Synthetic96PlateWithObjects	2	67.258437	60.400832	39	38	96	...	0.927387	0.777088	0.899936	0.161844	0.208367	0.561625	0.911701	0.350075	0.777088	9.988400
2	.png	8	GridImage	Synthetic96PlateWithObjects	3	67.774661	122.493213	39	99	98	...	0.931309	0.768554	0.903858	0.169159	0.220201	0.552949	0.915622	0.362673	0.768554	10.029955
3	.png	8	GridImage	Synthetic96PlateWithObjects	4	66.741818	615.873247	39	594	94	...	0.946995	0.776710	0.911701	0.173114	0.222997	0.560792	0.931309	0.370516	0.776710	9.446719
4	.png	8	GridImage	Synthetic96PlateWithObjects	5	67.689404	737.900864	39	714	98	...	0.939985	0.781367	0.914506	0.165666	0.212117	0.561625	0.919544	0.357918	0.781367	10.150514

5 rows × 179 columns

Summary

Prefab pipelines are the fastest path from plate image to results:

• ``HeavyOtsuPipeline`` — robust general-purpose detection with refinement

• ``RoundPeaksPipeline`` — lightweight peak-based detection for round colonies

• ``HeavyWatershedPipeline`` — for touching or clustered colonies

• ``FilamentousFungiPipeline`` — specialized for branching fungal morphology

• All prefabs support .apply(), .apply_and_measure(), .to_json(), etc.

Choose based on your organism, plate conditions, and desired accuracy. When a prefab is close but not quite right, use it as a starting point and customize the parameters.

Next up: Tutorial 9: Diagnosing Image Quality — assess plate quality before choosing a pipeline.