01 · Input
ND2 or TIFF
Use supported Nikon ND2 or TIFF microscopy data. Multidimensional ND2 selections are recorded explicitly.
BioNuclei · open scientific software
See the nuclei.
Measure what matters.
BioNuclei turns fluorescence microscopy into an inspectable nuclear analysis. Upload supported imaging data in the Analyzer; the released workflow segments nuclei, derives measurements, lets you explore the returned evidence, and packages a reproducible report.
Illustrative microscopy style view. Not benchmark evidence.
What it does
One focused workflow, now inspectable.
BioNuclei remains intentionally narrow: nuclear instance segmentation and quantitative measurements for fluorescence bioimaging, with an interactive review layer inspired by proven bioimage-analysis workflows.
02 · AI
Boundary U-Net
A convolutional segmentation model predicts background, nuclear interior, and boundary information.
03 · Result
Nuclei + measurements
Get labelled nuclear instances, geometric and intensity measurements, visual overlays, structured results, and provenance.
Interactive review
Do not stop at the mask.
A completed analysis can be explored in the Analyzer through a measurement-aware viewer, per-nucleus table, spatial ROI summary and quantitative heatmap controls.
Viewer
Inspect the analyzed overlay and segmentation mask before interpreting measurements.
Objects
Work with explicit per-nucleus instances and centroids rather than a single aggregate count.
Measurements
Review area, circularity, eccentricity, fluorescence intensity and centroid coordinates.
Spatial review
Draw a temporary ROI and inspect the nuclei and summary statistics inside it.
Prediction → instances → measurements → review.
The interactive layer does not replace the scientific pipeline. It exposes its outputs so a scientist can inspect what was actually computed.
Open the AnalyzerThe model
AI predicts. Scientific code measures.
The current scientific model is a compact Boundary U-Net. The adaptive layer can inspect the acquisition and route a validated workflow; it does not silently change the model weights.
Fluorescence2-D analysis plane
→
Boundary U-Net3 semantic classes
→
Instancesindividual nuclei
Input1-channel fluorescence for the current scientific model.
Output classesBackground · nuclear interior · boundary.
Post-processingSemantic predictions are converted into explicit instances.
MeasurementsArea, centroid, bounding box and derived reports.
No online training: the normal user-analysis path performs inference with the configured checkpoint. An uploaded image does not update model weights.
Adaptive analysis
Agents help the workflow adapt without replacing evidence.
Intelligent routing and specialist review can reduce manual setup while numerical results remain attributable to the segmentation and measurement code.
AGENT 01
Input reviewer
Reads format, dimensions, dtype, intensity statistics, and acquisition metadata.
AGENT 02
Workflow planner
Selects only an analysis profile that is actually available and validated.
AGENT 03
Quality reviewer
Flags measurable input problems and interprets segmentation evidence conservatively.
AGENT 04
Report writer
Explains what ran, what was measured, what warnings occurred, and what files were produced.
Scientific boundary: agents route and explain. The CNN predicts. Deterministic BioNuclei code computes measurements.
Your result
Evidence you can inspect.
A completed analysis is packaged so a scientist can inspect the image, explore object-level measurements, review spatial patterns, and trace the computation.
Visual evidence
Analyzed overlay, instance-labelled segmentation, and measurement-aware review.
Machine-readable evidence
segmentation_mask.tifindividual nuclear instances
overlay.tifvisual quality check
nuclei_analysis.csvper-nucleus measurements
results.json + provenance.jsonstructured result and lineage
Ready to analyze an image?
Open the Analyzer, choose an access mode, upload a supported ND2 or TIFF, inspect the returned evidence, download the report, and delete the transient job.
Open the Analyzer →