Research and educational proof of concept — not for diagnosis or patient care
Cell Bound

Transparent visual group analysis

See how the occupied-area estimate changes when one visual group is hidden.

Cell Bound groups visually similar image regions without pretending to identify cells. Inspect the overlay, quantify each group, then temporarily hide a group from the fixed-denominator image proxy.

PNG, JPEG, or WebP · up to 30 MB · file paste supported

01Designed for ordinary camera exports without a slide scanner.
02Measures occupied area with a fixed, visible denominator.
03Keeps the image in your browser; GPT‑5.6 sees derived metrics only.

Live proof of concept

Fruit-stand view

The full field—or a selected rectangular region—is the fixed denominator. Visual-group percentages and heuristic image compartments all use that same region.

No field
Choose an image to beginPNG, JPEG, or WebP · up to 30 MB

Local image workspace

Bring in one representative field

Upload, drop, or paste a de-identified PNG, JPEG, or WebP image. Maximum file size: 30 MB.

Analysis regionFull field
Enter exact region
Advanced exploratory setting

This recalculates every image compartment. It has no validated clinical cutoff and is provided only to explore the heuristic.

Segmentation and clustering run locally in this browser.Use a de-identified, pre-cropped representative field.

The fruit-stand approach

Similarity first. Names later—if ever.

From a distance, limes still look like a group even when each lime is slightly different. Cell Bound applies that operational idea to image patches: it estimates heuristic image compartments, groups recurring visual patterns, and keeps every number auditable.

⌁

Partition the field

Bright neutral space and low-nuclear-signal substrate are heuristic image compartments—not identified tissue classes. The remainder becomes occupied candidate area.

•••

Group by appearance

A seeded, deterministic clustering pass compares color, texture, darkness, and nuclear-signal features. It never assigns biological cell names.

−

Hide and inspect

Each ranked visual group reports its share of the selected ROI. Hiding removes that area from the numerator while the denominator stays fixed.

Why Cell Bound exists

Designed for ordinary camera exports without a slide scanner.

Bone-marrow cellularity is a visual skill built through repeated exposure and mentorship. This experiment explores a transparent educational image proxy for pathologists and trainees using ordinary microscope-camera exports.

✓
Local-firstUploaded pixels stay in the browser during segmentation and clustering.
✓
Human-reviewableThe overlay and every denominator remain visible instead of hiding the calculation.
✓
Intentionally boundedNo diagnosis, no cell-type identification, and no claim of clinical validation.