Microscopy analysis workflows

A visual, notebook-based guide to quantitative microscopy image analysis—from raw images and metadata to segmented objects, interpretable measurements, learned representations and analysis-ready tables.

Four run modes

Each route keeps the scientific hand-off explicit: image arrays become masks, label images or identity-preserving tables that the next stage can consume.

Starting data	Package and main call chain	Hand-off	Example	Walkthrough
2D fluorescence image or 3D stack	nuclear-imaging-core: segment_frame() → region_feature_table()	Label image, foreground mask and object table	01_segment_images.py	01_images_and_segmentation.ipynb
Nuclear and spheroid label images	nuclear-spheroid-analysis: measure_spheroid_system() → analyze_radial_2d/3d()	Nuclear, spheroid and radial tables	02_spheroid_radial_analysis.py	02_object_and_radial_features.ipynb
Ordered nuclear crops	nuclear-imaging-core.graph: dense-region segmentation → graph construction → tracking	Node, edge, graph-summary and tracking tables	03_spatial_graph_analysis.py	03_spatial_graphs.ipynb
Normalized nuclear crops and an existing split manifest	nuclear-vae-embeddings: fit_vae() → reconstruct_images() → latent_feature_table()	Reconstructions and learned-feature table	04_vae_representations.py	04_vae_representations.ipynb

All four scripts keep imports together at the head and accept caller-owned paths. They write runtime artifacts only to the requested output directory. See the examples guide for their input contracts.

Segmentation tasks

Brightfield spheroids	RGB transmitted-light spheroids
Raw field, connected labels and morphology-adjusted masks	Original RGB image, HSV-derived signal, object labels and seeds

Confocal z-stack nuclei (2D projection)	Fluorescent nuclear masks and intensity strata
DAPI projection, StarDist labels, watershed seeds and separated nuclei	Nuclear crops, object masks and within-mask intensity partitions

Dense intranuclear regions form a second segmentation scale: local chromatin domains become typed nodes and edges while remaining linked to the parent nucleus.

Follow the workflow

The notebooks are ordered so that every stage introduces one methodological decision, shows the corresponding Python interface and then presents a representative visual result.

Notebook	Focus	Result
00_workflow_overview.ipynb	Analysis design and data flow	End-to-end map
01_images_and_segmentation.ipynb	Image inspection and 2D/3D segmentation	Modality-specific label checks
02_object_and_radial_features.ipynb	Morphology, intensity, texture and radial measurements	Feature interpretation
03_spatial_graphs.ipynb	Dense-region graphs and temporal node tracking	Spatial representation
04_vae_representations.ipynb	VAE/CVAE embeddings of nuclear crops	Input/reconstruction comparison
05_feature_tables_and_figures.ipynb	Filtering, joins and condition-level analysis	Heatmaps and classification

Start with the workflow overview.

Method families

• Confocal stacks, multichannel fluorescence, RGB and brightfield inputs
• StarDist and threshold-based nucleus segmentation
• Two- and three-dimensional morphology and intensity measurements
• Texture, boundary curvature and nuclear/spheroid radial profiles
• Graph representations of dense intranuclear regions
• VAE and conditional VAE image embeddings
• Feature-table filtering, correlation pruning, clustering and visualization

All geometric thresholds in the examples are explicit pixel values. Adjust them in the JSON configuration files to match the image scale and assay.

Installation

Clone the workflow and method repositories into the same parent directory, then install them in editable mode:

python -m pip install -e ../nuclear-imaging-core
python -m pip install -e ../nuclear-table-tools
python -m pip install -e ../nuclear-spheroid-analysis
python -m pip install -e ../nuclear-vae-embeddings
python -m pip install -e .

Open the notebooks with JupyterLab:

jupyter lab notebooks/

Representation and analysis outputs

VAE reconstruction quality	Analysis-ready feature matrix

The method packages can also be used independently in experiment-specific scripts and notebooks.