Architecture

This repo is a napari plugin organized around a Qt widget and a set of tab modules.

Entry points

• src/senoquant/napari.yaml declares the napari widget and reader.
• src/senoquant/_widget.py defines SenoQuantWidget and registers the tabbed UI.
• src/senoquant/_reader.py exposes the reader entrypoint.

UI structure

The main widget (SenoQuantWidget) composes eight tabs:

• Segmentation (src/senoquant/tabs/segmentation).
• Spots (src/senoquant/tabs/spots).
• Prediction (src/senoquant/tabs/prediction).
• SenNet Portal (src/senoquant/tabs/sennet_portal).
• Quantification (src/senoquant/tabs/quantification).
• Visualization (src/senoquant/tabs/visualization).
• Batch (src/senoquant/tabs/batch).
• Settings (src/senoquant/tabs/settings).

Each tab follows a frontend/backend split:

• frontend.py builds the Qt widgets and handles UI events.
• backend.py performs the model discovery or processing logic.
• Segmentation additionally uses segmentation/_frontend/ mixins to keep UI code split across smaller modules.

Prediction-specific note:

• The tab-level UI is fixed (Select model, Model interface, Run), and each prediction model contributes its own Qt widget through SenoQuantPredictionModel.build_widget(...).

Reader pipeline

The reader implementation lives in src/senoquant/reader/core.py and uses a hybrid BioIO strategy. The reader:

• Prefers fast format-specific BioIO plugins for pixel loading.
• Uses bioio-bioformats for metadata extraction when available.
• Falls back to the data reader for metadata when BioFormats metadata open fails.
• Uses dask_tiles=True when BioFormats is used for pixel loading.
• Extracts channel colors from OME metadata when available.
• Falls back to a default colormap cycle when all channels have the same color.
• Iterates scenes and channels to create napari layers.

Batch pipeline

Batch processing is orchestrated by BatchBackend in src/senoquant/tabs/batch/backend.py. The flow is:

1. Enumerate input files and resolve channel indices.
2. Run segmentation (nuclear/cytoplasmic) as configured.
3. Run spot detection per selected channels, then apply optional diameter-based spot filtering.
4. Run quantification using a lightweight viewer shim.
5. Write masks and quantification outputs to disk.
6. Persist senoquant_settings.json in the batch output root with the effective batch_job configuration.

Prediction pipeline

Prediction runs are orchestrated by PredictionTab + PredictionBackend:

1. Discover model folders under src/senoquant/tabs/prediction/models/.
2. Load the selected model class from <model_name>/model.py.
3. Build model-defined UI in the Model interface box.
4. Collect model widget settings and run model code in a background thread.
5. Normalize model output into napari layer specs and add layers to the viewer.
6. Append run metadata (task="prediction", runner name/type, settings).

Settings storage

The Settings tab uses SettingsBackend to read/write unified senoquant.settings JSON bundles. These bundles can include:

• tab_settings payloads for tab-level settings snapshots.
• batch_job payloads compatible with batch config serialization.
• feature_settings and segmentation_runs payloads produced by quantification exports.

Cross-tab settings orchestration

SenoQuantWidget instantiates all tab widgets. Settings currently receives Segmentation, Spots, and Batch tab references and uses them to:

• Export segmentation and spots configuration into tab_settings.
• Export batch state into batch_job.
• Restore those states when loading a bundle.

Prediction, Quantification, and Visualization settings are currently not restored by the Settings tab.