ImFusion WebAppKit

The ImFusion WebAppKit serves Python imaging algorithms in a medical web viewer over HTTP, so you can demo them to colleagues and clinicians without writing any web code. You register a function, the kit builds the matching button and controls, and whatever you return shows up in the viewer.

The kit already includes the viewer, the HTTP server, and a session model that isolates each browser connection. What you write is the imaging logic, which runs on the ImFusion Python SDK thread rather than in the browser.

This is an early prototype, so interfaces may still change between versions. The kit itself is MIT-licensed, but it runs on the ImFusion SDK, which is free for non-commercial use and needs a commercial licence otherwise.

Creating an app

Annotate a function and give it the name you want on the button, and WebAppKit builds the matching controls and runs the call on the ImFusion SDK thread. Argument types pick the controls, so a float with bounds arrives as a slider and an image argument as a dataset selector. You can also expose an existing ImFusion algorithm by name.

import numpy as np
import imfusion

from imfusion_webappkit import FloatParameter, ImFusionWebApp

app = ImFusionWebApp(title="Image Tools")


@app.register(
    "Threshold",
    parameters=[
        FloatParameter(
            "threshold", default=100.0, minimum=0.0, maximum=1000.0,
        ),
    ],
)
def apply_threshold(
    imageset: imfusion.SharedImageSet,
    *,
    threshold: float,
) -> imfusion.SharedImageSet:
    image = imageset[0]
    label = imfusion.SharedImage((image.numpy() >= threshold).astype(np.uint8))
    label.image_to_world_matrix = image.image_to_world_matrix
    label.spacing = image.spacing

    mask = imfusion.SharedImageSet()
    mask.add(label)
    mask.modality = imfusion.Data.Modality.LABEL
    return mask


app.run()

Open http://127.0.0.1:8000, load an image, select it, and run the action.

What's included

The imaging work stays in Python, while the kit owns the parts that usually force you into a frontend project.

The viewer. MPR, 3D, and 2D views with the display options and interactions people already know from ImFusion Suite. Layout and which views are visible are set from Python (ViewLayout, LayoutConfig).

The data model. Add, rename, or clear data through app.data_model, and the browser follows without any message passing of your own.

A session per visitor. Every browser connection gets its own data and workflow state, so several people can open the app at once.

Your branding. Logo, colors, panel widths, an About dialog, and bundled sample datasets can be configured (BrandingConfig, ThemeConfig).

A command line. doctor checks the environment, demo runs a sample app, init scaffolds a project, and record writes a snapshot you can host with no backend.

Guided workflows

If a study has to run in a fixed order, define a workflow as a list of steps. The kit handles the UI and keeps state per session, including progress and cancellation. A reviewer can reject a result and step back without restarting.

The built-in steps cover loading and input roles, parameters, processing, brush correction, annotations, validation, and export, and a CustomStep composes a panel of your own when none of them fits.

app.set_workflow([
    MessageStep("Welcome", "Load, segment, correct, export."),
    InputSelectionStep("Select Image", inputs=[image]),
    ParameterStep("Configure", parameters=[threshold]),
    ProcessingStep("Segment", callback=threshold_segment),
    BrushStep("Correct Segmentation", radius_mm=5.0),
    SegmentationSummaryStep(label_map_from="process"),
    ValidationStep("Review", "Is this acceptable?"),
    ExportStep("Export Segmentation"),
])

The snippet is the segmentation example, with eight steps in the order they run and a validation step that can send the operator back to the brush. SegmentationSummaryStep is not one of the built-ins: the example defines it as a CustomStep, which is how a workflow gets a panel the kit does not ship.

The screenshot is a different workflow from the snippet above, where two steps assign fixed and moving images by role and the registered result then lands in the data model.

Start from a template

imfusion-webappkit init creates a runnable project. The default is a single action; the other flags pick a different shape.

--template simple. A button, the parameters you declared, and your function behind them.

--template workflow. Load, process, brush-correct, and a review the operator has to accept before export.

--template monai. The spleen bundle from the MONAI Model Zoo, run on the sample CT. The download, tensor layout, and coordinate conversions are already in the template.

--template chat. A dataset-aware chat panel with a placeholder reply function, ready to connect to your own model provider.

Get started

The kit needs Python 3.10 or newer, uv, and valid ImFusion Suite and Web SDK licenses. Releases are published to PyPI, but the ImFusion SDK underneath them comes from the ImFusion index instead, so an install has to name both:

uv pip install imfusion-webappkit --extra-index-url https://pypi.imfusion.com/simple
imfusion-webappkit demo

From a checkout of this repository the index is already configured in pyproject.toml, so uv sync resolves everything on its own:

uv sync
uv run imfusion-webappkit demo

That starts the packaged demo with a small bundled NIfTI image at http://127.0.0.1:8000. Add --workflow to launch the guided, multi-step demo instead.

The checkout includes the compiled browser client, so Node.js and npm are not needed to run it. They are only required to modify the frontend; see the development guide.

To scaffold your own project:

uv run imfusion-webappkit init my-demo
cd my-demo
uv sync
uv run python app.py

The generated project runs as soon as it is created, with app.py holding the application, algorithm.py the processing function you replace, and AGENTS.md the session, threading, and geometry conventions that coding agents need. Pass --theme gray for the gray theme instead of the default dark one, and see the command-line documentation for the other templates and flags.

Documentation

• User guide
• Getting started
• Command-line tools
• API reference
• Runnable examples
• Development guide

If the kit does not fit how you work, or you need a commercial ImFusion SDK licence, email info@imfusion.com.