MTZ² 2D, 3D and 4D microscopy in your browser
New · AI Assistant that executes the analysis

2D, 3D and 4D microscopy in your browser · free

From raw signal to measurement,
right in your browser

TIFF, OME-TIFF, CZI and LIF confocal stacks analyzed in your browser: automated segmentation with parameters measured on each image, an AI assistant that executes the analysis and cites measurements, Richardson-Lucy deconvolution, machine learning trained on your own examples, and the FAIR Capsule for peer reviewers. No paid licenses and without uploading a single image: each algorithm calibrated against known references, every metric with its provenance, and whatever cannot be measured stated with a dash rather than an invented number.

0volume error on reference sphere
0nuclei automatically segmented by MTZ² in this image alone
0bytes of your images sent to any server
Scroll: the image is analyzed

01 · Behind the image

Not a mock-up: an authentic specimen

A three-channel single-page tissue fluorescence image: nuclei in blue, labeled cells in red and green. The background image measures 1920 × 1920 pixels (original: 3008 × 3008), and everything rendered on top as you scroll was computed by MTZ² upon loading, using the exact same application you are about to run.

The image carries no calibration metadata, so measurements are reported in pixels. Converting to microns without a validated voxel size would mean inventing a number—the single thing MTZ² strictly forbids.

02 · Color fidelity

Every channel in its true color, none synthesized

An RGB TIFF is parsed into three distinct channels and rendered using full-dynamic-range pure primaries: the only colors that, when additively combined, reconstruct the raw image exactly. That is what scrolls in the background: red, green, blue, and all three merged.

This was an authentic bug reported with this very image: 97.6 % of its signal pixels are blue-dominant, yet the original viewer rendered them in orange and turquoise. The color fidelity test suite now verifies rendering pixel by pixel.

03 · Thresholding

Differentiating signal from background, rigorously computed

The threshold is determined by Otsu’s method: the intensity value that maximizes between-class variance between foreground and background. On the blue channel of this image, Otsu yields 68, visualized in the background sweep illuminating signal above cutoff.

Blue channel · height: square root of bin count cutoff = 68 · Otsu

Under uneven illumination, a single global cutoff fails. In the reference benchmark, global thresholding detects 48 objects where 16 disks exist, whereas regional adaptive thresholding recovers all 16.

04 · Segmented objects

646 nuclei, identified individually

This is the automated segmentation MTZ² performs on opening the image, using parameters empirically measured from its signal. The outlines shown are computed directly, with zero manual retouching.

  • Smoothing of 3 px. The minimal scale reaching 95% compactness across a 9-scale sweep: suppresses chromatin granulation without merging adjacent nuclei.
  • Regional thresholding: local Otsu in each of the 13 × 13 tiles (152 px wide), ensuring dim peripheral areas retain proper detection.
  • Expected diameter of 51 px, measured directly from the raw signal, with watershed declumping of touching nuclei.
  • Size gating from 5% to 400% of nominal area (101 to 8042 px²): filtering 849 candidate regions down to 646 verified objects.

Without smoothing, the same algorithm shatters chromatin into 5145 fragments, yielding 893 artifactual "objects" where 646 true nuclei reside. Size filtering discards 190 undersized regions and 13 oversized regions (some of which are genuinely large nuclei). Discarded items are reported alongside successes: a method that reports only its hits cannot be scientifically evaluated.

05 · Dimensionality

A single plane is treated as a plane

This image contains a single 2D plane, and MTZ² renders it faithfully as such. Given a Z-stack, it traces it as a GPU-accelerated 3D volume; given a time series, it plays it back in 4D. The dimensional selector informs you at all times of the raw geometry and its active viewport representation.

Unsupported modes are not hidden: they remain visible in disabled state, explaining why. Enabling "3D" on a single slice would not just yield a poor view; it would produce a blank screen indistinguishable from a file loading error.

06 · Time & Tracking

Tracking trajectories while preserving object identity

Particle tracking solves assignment via global linear assignment optimization, not greedy nearest neighbors: when two trajectories cross, each retains its correct identity. Benchmark against a known Brownian motion diffusion process:

measured
0.16795
theoretical
0.1667

Apparent diffusion coefficient in µm²/s, verified in the tracking test suite. The suite confirms that trajectory identity survives crossover events and recovers particles that disappear for up to two consecutive frames.

07 · Your data privacy

Your images never leave your computer

MTZ² runs entirely within your browser engine. Files are loaded directly from your local disk and processed in memory: no uploads, no server copies, and no cloud processing. The authentication server stores only minimal account metadata: name, email, Google user ID, and sign-in timestamps. If you enable the AI assistant, only textual prompts and measurements are sent to your chosen LLM provider—never raw pixel arrays. The only exception is by your explicit request: during a Live Sync session, your browser streams the image peer-to-peer and encrypted directly to invited colleagues. Only if strict NAT firewalls prevent direct P2P (and you authorize relaying), does encrypted traffic transit the MTZ² server, which purges it upon session termination.

Median nucleus area1769 px²
Area in µm²— · uncalibrated
Volume— · single slice

Verifiable in DevTools: the Content Security Policy permits connections exclusively to its own server (app shell & auth), your configured AI provider (text queries only), and AlphaFold DB (protein accession lookup). The browser blocks all other network traffic, and none of these transfer pixel data. Live Sync adds, solely during an active peer review session, direct WebRTC encrypted channels to your invited peers.

In motion

Your datasets, rendered live by MTZ²

Two authentic acquisitions, opened in the browser with zero pre-conversion: a whole mouse brain from a Zeiss lightsheet and a 3-channel Leica immunofluorescence leading to imaging cytometry. Every frame was rendered by the application using a standard laptop GPU; only cross-fades and text captions were added.

Whole mouse brain in 3D. A 4.5 GB, 859 M voxel Zeiss Lightsheet CZI, rendered via GPU raymarching with Fire colormap and calibrated scale bar.
From imaging to cytometry. A 3-channel Leica LIF immunofluorescence: 2703 nuclei segmented upon loading, with green vs. red intensity plotted per cell in Vantage, density-colored with interactive quadrant gating.

Capabilities

Nine quantitative answers on a single image

Not nine disparate tools: a single coherent model of the specimen subjected to nine scientific questions. Nuclei counts, channel intensities, and colocalization stem from the same validated segmentation, sharing identical calibration and provenance.

01

Surfaces & segmentation

Parameters measured on each image upon loading (smoothing, diameter, regional thresholding), watershed separation of touching objects, and size gating. Validated against 29,438 expert-annotated nuclei.

02

Spots & synapses

Difference of Gaussians with sub-voxel parabolic refinement, local background subtraction, and a machine learning filter trained on points you click.

03

Filaments

Geodesic centerlines and Sholl analysis. If the structure is fragmented, it explicitly reports the discontinuity rather than hallucinating paths.

04

4D Tracking

Global linear assignment, birth and death events, gap closing, mean squared displacement, and apparent diffusion coefficients.

05

Colocalization

Pearson, Spearman, and Manders coefficients with Costes automated thresholding. If a channel is flat, it returns an explanation instead of an undefined number.

06

Brush pixel classifier

Random forest trained by painting directly on slices: up to eight classes and thirteen multiscale descriptors, retrained in milliseconds.

07

Contours & ROI catalogue

Hand-drawn regions and editable objects, each tagged with its exact origin: measured, derived, or user-edited.

08

3D metric measurements

A caliper ruler that snaps to structural surfaces rather than bounding boxes, and scale bars computed from optical camera and calibration metadata.

⏱

Provenance, above all

Dataset SHA-256 hash, reproducible Methods paragraph with formal citations, and in-browser instrument validation certificate.

Discovered during calibration

Fifteen edge cases uncovered through calibration

None of these are embellishments: each is documented with its regression test. Almost all represent identified and resolved algorithmic bugs. We publish them openly because an analytical instrument whose flaws remain concealed cannot be scientifically audited.

−51.0 %

The sphere lost half its volume

An incomplete Marching Cubes lookup table underestimated sphere volume by 51.0 %. Today error is 0.1 %: 2141.47 µm³ vs. 2144.66 µm³ theoretical.

0.75

An invented colocalization value

When computation failed, the Pearson coefficient returned a hardcoded 0.75, indistinguishable from a true measurement. Today it returns null, and the UI explains why.

48 / 16

A single cutoff is insufficient

On uneven illumination, global thresholding counts 48 objects where 16 disks exist. Regional thresholding finds all 16, declaring the tile grid.

0 µm³

CSV exported zero volume

An unassigned variable caused every CSV export to output Volume = 0. Nobody noticed initially because zero resembled a valid measurement.

3.213

Confidence interval was too narrow

For small samples, the 95% confidence interval must use Student’s t distribution: 3.213 to 6.787, not 3.518 to 6.482 as Gaussian z approximation produced.

51.5 %

Data leakage stress test

With randomized labels, the classifier memorizes training data but out-of-bag accuracy drops to 51.5 % (pure chance). Exactly what an honest model must do.

+320 %

Deconvolution that never deconvolved

The button displayed "Completed (+320 % peak SNR)" after 300 ms without altering a single pixel. Today it executes true Richardson-Lucy: correlation on a noisy phantom rises from 0.21 to 0.74, reporting only measured metrics.

0.125

3D volume downsampling erased detail

Volume textures were capped at 256 voxels per axis: fine detail correlation dropped to 0.125 and only 38 % of bright spots were visible. At native resolution: 0.947 and 97 %.

0 clicks

A classifier that did not learn

The spot ML filter announced "Model trained" without fitting anything. Today it trains a 20-tree random forest from points you mark, reporting genuine out-of-bag accuracy.

124

Capsule containing dummy data

The FAIR Capsule populated missing fields with mock figures: 124 cells, 642 spots, fixed SHA-256 hash. Today it contains solely active session data, and missing measurements display as "—".

2.4 GB

Large files crashed on load

The reader loaded whole files into memory, hitting Chrome’s 2 GB ArrayBuffer ceiling: a 2.4 GB OME-TIFF could not open. Today it streams plane-by-plane with no size cap, opening in ~15 s.

4 bytes

Zeiss and Leica parsing bugs

The CZI parser searched directories 4 bytes off; LIF read headers from the wrong offset yielding 512 × 512 blank frames. Rewritten to vendor specs, they match reference readers bit-for-bit: 5 CZI datasets (including a 4.5 GB mosaic) and 2 LIF projects.

0.088

Stacked nuclei fused along Z

In 3D, smoothing used the same voxel kernel in Z as in XY. With 1 µm Z step and 0.2 µm XY pixel, Z smoothing was 5× oversized, fusing stacked nuclei (F1 0.088). With isotropic physical sigma in microns across all 3 axes, F1 reached 0.900 (mean: 0.67 → 0.94).

+88

Spot local background contamination

Background annulus fell into the spot’s own point spread tail: up to +88 levels excess, underestimating net intensity. Sized between 2–3 radii using median estimator, mean error dropped from 34 to 2.5 levels.

12.3 GB

Whole organ lightsheet froze browser

Opening an 859 M voxel mouse brain automatically triggered segmentation: > 20 min run, 12.3 GB memory spike and 17.9 s freeze. Now, massive datasets run no heavy tasks unprompted and build 3D textures in chunks: ~4 GB and UI latency < 2 s.

5145

This page’s background image

Without smoothing, segmentation breaks chromatin into 5145 fragments. That is why smoothing is measured individually: here, 3 px, resolving 646 genuine nuclei.

The method

Zero proprietary black boxes where peer-reviewed science exists

A proprietary black-box algorithm that cannot be audited is useless for peer-reviewed research. Implementations built upon Otsu (1979), Lorensen & Cline (1987), Sholl (1953), Manders (1993), Costes (2004), or Breiman (2001) allow full validation. No one owns mathematical equations; what matters is numerical precision and fidelity. Below is the exact Otsu formulation sweeping the background canvas.

σ²B(t) = ω0(t) · ω1(t) · [ μ0(t) − μ1(t) ]² → t* = argmax σ²B
  1. Pixel proportions on each side: ω₀ and ω₁ are the fractions of pixels below and above threshold t. A split placing nearly all pixels on one side yields a negligible product.
  2. Separation distance: μ₀ and μ₁ are the mean intensities of both populations. A rigorous threshold separates two distinct distributions, not one distribution into halves.
  3. Between-class variance: the product of both terms, rewarding balanced class partitioning while maximizing distance between population means.
  4. Optimal threshold: all 256 intensity levels are evaluated to identify the maximum. In case of ties, the midpoint is selected to prevent falling into histogram minima.

Machine Learning

Classifiers that learn directly from your annotations

With photobleaching, depth attenuation, or autofluorescence (lipofuscin, collagen, antibody precipitates), simple intensity thresholding struggles to isolate genuine signal. MTZ² provides two Random Forest classifiers (Breiman, 2001) that learn interactively from your manual marks, right inside your browser.

🎯

Spots: click and train

Following detection, click true spots and false positives (debris, noise, autofluorescent granules). A 20-tree forest learns from 8 morphological descriptors (radius, peak, background, SNR, FWHM, DoG response…), allowing interactive probability filtering.

🖌️

Pixels: paint and segment

Paint interactive strokes across slices for up to eight classes. A 40-tree forest fits 13 multiscale voxel features (intensity, Gaussian scales, gradients, DoG, local variance) and segments the full volume, generating quantitative surfaces for each class.

🛡️

A model that does not deceive

Every training session reports out-of-bag accuracy and alerts if class support is sparse. With shuffled labels, accuracy drops to 51.5 % (random guess). Retrained in milliseconds in memory: zero voxels leave your device.

Innovations for Research & Medicine

Nine major innovations, backed by benchmark proofs

Engineered for scientists, biomedical researchers, and clinical investigators. Every single feature operates right now in your browser, without transmitting your images anywhere.

🤖 EXPERT AGENT

AI Assistant that executes analysis

Instruct it in plain English: "segment the nuclei on the DAPI channel, count the γ-H2AX foci, and draft the Methods section". The assistant loads, segments, detects, and measures using the exact same underlying functions as the UI buttons (18 tools), citing exclusively measured numbers and consulting the Manual. It reasons as a specialist in microscopy, pathology, microbiology, and genetics. Compatible with Claude, Gemini, OpenAI, or local models (Ollama) using your private key.

For the scientist: reduces repetitive chores while keeping every step transparent. Only text prompts reach providers; destructive operations require confirmation.
✨ ON-LOAD DETECTION

Validated automated segmentation

Upon opening an image, MTZ² measures smoothing scale, expected object diameter, and regional thresholding directly from the data, performing autonomous segmentation. Validated across 29,438 expert-annotated nuclei: F1 score of 0.899 (BBBC039) and 0.790 (DSB 2018). An empty field is reported as empty rather than hallucinating objects.

For the scientist: an immediate, reproducible baseline in seconds, with measured parameters displayed openly for audit.
⚡ OPTICAL RESTORATION

Richardson-Lucy deconvolution

From emission wavelength, numerical aperture, and refractive index, it models theoretical Gaussian optical PSFs (Zhang et al., 2007) and restores 2D slices or 3D volumes asynchronously in web workers. The result creates a new channel—raw data remains untouched. In a Poisson-noise phantom, correlation to true ground truth jumps from 0.21 to 0.74.

For the scientist: enhanced contrast near diffraction limits, with automated Nyquist undersampling warnings and full Methods documentation.
📦 PEER REVIEW EVIDENCE

Self-contained FAIR Capsule

A single standalone .html file opening completely offline: SHA-256 voxel checksum, calibration, rendered figure, Methods paragraph in text and LaTeX, tabular CSV with full provenance, 3D OBJ mesh, and SVG plots. Derived exclusively from session data; unmeasured attributes display as "—". Contains zero proprietary image pixels.

For the scientist: reviewers verify calculations and methodology instantly without software installations or licensing barriers.
🧬 CELL TO MOLECULE

AlphaFold DB structures

Fetches predicted protein structures via UniProt accession, computes per-residue pLDDT confidence scores, and renders Cα backbone traces at fluorescent spot centroids. Proteins measure nanometers (orders below optical resolution): rendered magnified with MTZ² explicitly declaring scale factors.

For the scientist: direct structural context for targets seen under microscopy; designated as illustrative rather than molecular colocalization.
🧊 VERIFIED FIDELITY

What you see is what is there

2D planes render pixel-identical to source files, while 3D volumes are raymarched at native resolution with > 8-bit depth up to hardware limits. Compared to earlier iterations, fine detail resemblance improved from 0.125 to 0.947, and visible bright spot recovery rose from 38 % to 97 %.

For the scientist: single-voxel filaments and faint puncta never vanish when transitioning into 3D space.
👥 LIVE PEER REVIEW

Live Sync: shared 3D specimen review with colleagues

Share an invite link (via email or chat) to inspect datasets together: host and up to 25 remote participants. Camera motion synchronizes in real time with laser pointers, named 3D pins, and shared notes. Peers lacking the file receive it directly from your browser, encrypted and with progress telemetry. SHA-256 hashes ensure everyone inspects the exact same voxel matrix. Verified through 99 cross-browser automated tests.

For the scientist: lab meetings, seminars, and second opinions on volumetric specimens within two minutes. Transferred peer replicas are purged upon closing.
🧪 IMAGING CYTOMETRY

Flow cytometry on segmented cell populations

Vantage computes mean intensity per cell across channels, rendering interactive cytometric scatter plots: density heatmaps, log scales, and quadrant, rectangular, or polygonal gates. Subpopulations are isolated, statistically tested (Welch, Mann-Whitney, Kruskal-Wallis with Holm, Pearson, Spearman), and highlighted in 3D tissue context. Validated on a 220-cell benchmark: all 4 quadrants match cell-for-cell.

For the scientist: marker-based immunophenotyping preserving spatial tissue coordinates, with export-ready tables for R, Python, and spreadsheets.
🌐 OPEN FORMAT

OME-Zarr: multiscale pyramid arrays

Open a local .zarr folder (drag and drop) or stream directly from public OME-NGFF repositories: each resolution level maps as a series with calibration, channel names, colors, and display windows. Supports Zarr v2 and v3 chunk streaming (Blosc, Zstd, Gzip, Zlib, LZ4, sharded stores). Validated identical to reference libraries across 25 benchmark datasets.

For the scientist: modern bioimaging archive format streamed seamlessly without prior file conversions.

Comparative analysis

Comparison: MTZ² vs. Desktop Bioimaging Suites

MTZ² brings into your browser core workflows of the gold-standard commercial 3D/4D microscopy suite—surfaces, spots, filaments, 4D tracking, colocalization, machine learning classification, linked statistical plots, and dendritic spines—while adding modern innovations marked as new. The table benchmarks capabilities: for massive terabyte-scale datasets or specialized niche modules, dedicated workstation software remains appropriate.

Capability Commercial desktop suite Open-source desktop software MTZ² (Browser)
Installation Desktop application (Windows & macOS) with license key Java desktop application; user-managed plugins Zero install: opens directly in the browser
Licensing & cost Commercial license (quote upon request) Free (open source) Free for scientific research, via Google account
New AI Assistant No conversational agent executing analysis Handwritten macros and scripts Autonomous agent executing analysis with 18 tools, citing measured metrics (Claude, Gemini, OpenAI, Ollama)
New On-load segmentation Manual parameter wizard Deep learning plugins with pretrained weights Autonomous signal measurement: F1 0.899 (BBBC039) and 0.790 (DSB 2018) across 29,438 nuclei
New Deconvolution Optional deconvolution module, license-dependent Deconvolution plugins Richardson-Lucy 2D/3D with theoretical Gaussian PSF, outputting to a new channel with measured metrics
New Display fidelity GPU volume rendering 3D viewer and multiresolution viewer 2D bit-identical rendering (RGB) and native resolution 3D volume (>8-bit depth), validated against raw data; loads microscope acquisition colors, display windows, and scientific colormaps (Fire, Viridis)
New Reviewer evidence CSV / Excel export; proprietary files require free vendor viewer Separate TIFF, CSV and macro scripts FAIR Capsule: single .html with SHA-256, figure, Methods, CSV, and 3D mesh, viewable offline with zero install
New Machine learning Object ML classifier, license-dependent Pixel classification plugins Random Forests trained on your clicks (spots: 8 descriptors) and painted strokes (pixels: 13 descriptors, 8 classes) with out-of-bag accuracy
New Molecular structure Not included Not included (separate structural biology software) AlphaFold DB: predicted structures with pLDDT scores, rendered magnified at spot centroids
New Manual & Recipes Vendor user manuals and video tutorials Community wiki and forums Full scientific and biomedical manual in English & Spanish, with protocols, recipes, and references; searchable by AI assistant
New Microscopy formats Imports vendor formats and converts to proprietary file format Import libraries: >150 formats supported Direct in-browser reading of TIFF, OME-TIFF, BigTIFF, LSM, Zeiss CZI (raw/Zstd, scenes, mosaics, lightsheet dual illumination), Leica LIF, and OME-Zarr (v2/v3, local/remote)
New Large files Proprietary multiresolution file format for multi-GB datasets Virtual stacks and multiresolution viewers Chunked plane-by-plane streaming without file size ceiling, chunked block processing up to allocated memory (1.6 GB default, expandable to 8 GB); beyond this, transparent XY downsampling. Responsive 3D textures with zero UI lockup
Data privacy Local workstation execution Local workstation execution In-browser execution: 0 bytes of pixel data sent to external networks; shared exclusively when initiating Live Sync
Analysis recipes Batch processing via project manager, license-dependent Macros and scripting Open JSON .mtzwf recipes with SHA-256 checksum validation upon import
Confined detection Spots and filaments inside surface boundaries User-generated binary mask workflow Confinement of spots and tracks to selected surface masks
Channel masking Masking channels inside or outside surface objects Image calculator applied to duplicate arrays Non-destructive channel generation with masked signals and intensity fraction statistics
Statistical plotting (Vantage) Statistical module, license-dependent Basic plotting; typically exported to external statistics package Vantage 1D, 2D, temporal, and spatial plots: Tukey boxplots, violin distributions, histograms, 95% CI regression, hypothesis tests, bi-directional 3D linking, and vector SVG export
New Imaging cytometry Per-object statistics and intensities, module-dependent Measurements per object; cytometry handled in external tools Density scatter cytometry on log scales with quadrant, rectangle, and polygon gates; isolated populations linked directly back to 3D
New Peer review Exporting static scenes or video screen captures Sharing dataset files and macro code Live Sync: real-time camera sync, laser pointer, 3D pins, and collaborative notes for up to 25 peers; encrypted P2P image transfer with SHA-256 verification
Dendritic spines Filament module with spine detection Neurite tracing plugins Spine neck/head diameter via FWHM and maximal inscribed sphere, with 4-class morphological categorization
Workspaces Integrated project and batch manager Independent window instances Arena Mode and 7 specialized functional workspaces
Provenance & Methods History log of creation parameters Macro recorder Peer-reviewed Methods paragraph with citations, dataset SHA-256 hash, and in-browser geometric phantom validation

Comparison compiled from public documentation of reference instruments in September 2026. Third-party features, modules, and pricing vary by release and licensing tier: consult respective vendors. "New": introduced in this MTZ² release. CZI and LIF parsers validated bit-for-bit against czifile and readlif with test benchmarks and authentic acquisitions (5 CZI datasets and 2 LIF projects).

Scientific outputs

Ten research-ready deliverables

A measurement that cannot be integrated into a lab notebook or publication is useless. Every single deliverable is exported with its attached provenance: which dataset, which calibration, and which numerical parameters produced each metric.

📦

FAIR Capsule (.html)

Self-contained evidence for reviewers: SHA-256 hash, figure, Methods, CSV, and mesh, with zero installation.

📕

Methods Paragraph

Written in academic English with citations: lists only executed steps with exact parameters. Also drafted by AI assistant.

📊

CSV Tables

Object by object, with descriptive statistics, cell-by-cell counts, and full acquisition provenance.

⚡

.mtzwf Recipe

Executed workflow steps and parameters in open JSON, validated with SHA-256 hash upon import.

💾

Session state

Complete analysis state and parameters, enabling continuation or direct sharing with colleagues.

📈

Vantage SVG figures

Tukey boxplots, violins, histograms, scatter regressions, and gated cytometry plots in print-ready vector graphics.

🖨️

STL 3D mesh

Watertight and metric-calibrated, ready for 3D printing of segmented structures.

🧊

OBJ mesh

With surface normals, compatible with standard 3D rendering and scientific visualization tools.

📷

High-resolution snapshot

Exact viewport render with calibrated scale bar, formatted for publication figures and presentations.

✅

Validation certificate

Automated benchmark results measuring geometric phantoms of known ground truth in your browser.

No execution path shall ever produce a number indistinguishable by the user from an authentic physical measurement.

This principle governs every architectural choice. Unmeasured quantities display as "—"; modeled values are explicitly tagged in the UI, CSV, and provenance logs; and whenever an analytical step cannot be computed, the application discloses why. With every new algorithm, the mandatory question is: what does it return when it cannot measure? If the answer is an invented number, it is scientifically flawed.

Commitment & Services

Free software. Expert assistance when needed.

Quantitative microscopy analysis should be accessible to everyone: researchers, imaging facility specialists, educators, and students. That is why MTZ² is and will remain free—no credit card required and no usage quotas; a Google account suffices. The project is sustained by services requiring dedicated human expertise: technical consulting, custom deployment, training workshops, and voluntary patron contributions.

Unlimited free access

Full Platform

0 €/ forever

All quantitative analysis tools, running right in your browser with a standard free Google account.

  • Segmentation, spots, filaments, 4D tracking, and colocalization
  • AI assistant with your own API key or local model, deconvolution, and FAIR Capsule
  • Direct TIFF, OME-TIFF, LSM, CZI, and LIF loading without conversion
  • CSV, STL, and OBJ exports with complete provenance metadata
  • Scientific manual with step-by-step protocols and recipes
  • Your images never leave your computer, unless shared in a Live Sync review
Sign in No credit card required, never expires
★ Professional Support
Expert Support

Technical Assistance

99,99 €/ month

Direct consulting from the author to prepare datasets, optimize analytical parameters, and interpret MTZ² findings.

  • Audit of optical calibration and dataset metadata
  • Parameter optimization for segmentation, spots, and tracking
  • Scientific interpretation of results and boundary conditions
  • 24–48 business hour response time by email
  • No lock-in: cancel at any time
Subscribe to support No commitment: cancel whenever you wish
Custom Scope

Deployment & Training

On request/ custom scope

Tailoring MTZ² to your facility’s file formats and protocols, and training your research staff.

  • Training workshops for research groups and facilities
  • Hands-on analysis clinics with your own microscope files
  • Integration of custom acquisition formats and metadata
  • Benchmarking and validation with your instrument’s phantoms
  • In-person or remote video sessions
Inquire In-person or online, tailored to needs

Technical Patronage & Project Support

Voluntary contribution to the project, from 1 €

MTZ² is and will remain free. If it saves you time, helps you prepare a scientific publication figure, or you simply believe open bioimaging tools should flourish, you may support its maintenance and continuous development with any voluntary amount, coordinated directly via email.

Coordinate contribution (1 €) via email Direct and confidential communication with the author at imagine@intelligentia.dev

THANK YOU to all who use, share, and improve this tool: scientists, imaging technicians, educators, and students. Without that trust, this project would not exist.

Suggestions: send feedback or improvement ideas Development collaboration: image analysis, optics, or code

Test it yourself

The complete web application, documentation with step-by-step protocols, and interactive demo samples to test without uploading your own files. A Google account is all you need: no credit card, zero installation.

Sign in Read the manual