IMPACT-Seg-KonfAI

Fast and lightweight CLI for multimodal anatomical segmentation using IMPACT-Seg models within the KonfAI framework.

🧩 Overview

IMPACT-Seg-KonfAI is a lightweight command-line interface (CLI) for running IMPACT-Seg models through the KonfAI deep learning framework.

It provides a simple way to perform anatomical segmentation inference, evaluation, ensembling, and uncertainty estimation on medical image volumes.

The underlying IMPACT-Seg models are multimodal anatomical segmentation networks built around a 2.5D U-Net with a residual encoder. A single model segments CBCT, MR, and CT scans into a consistent label space, enabling cross-modality workflows without per-modality retraining.

Pretrained models are automatically downloaded from Hugging Face Hub.

🧠 Features

• ⚡ Fast inference powered by KonfAI
• 🤗 Automatic model download from Hugging Face
• 🧩 Multi-model ensembling and test-time augmentation (TTA)
• 🧠 Supports evaluation workflows with reference data, and uncertainty estimation without reference
• 🧾 Multi-format compatibility: supports all major medical image formats handled by ITK

🗂️ Available models

Model identifiers are resolved dynamically from the VBoussot/ImpactSeg repository and passed as the first positional argument.

The body model predicts 11 anatomical labels spanning soft tissues, cavities, bones, and central structures:

🚀 Installation

From PyPI:

python -m pip install impact-seg-konfai

From source:

git clone https://github.com/fideus-labs/KonfAI.git
cd KonfAI
# konfai and konfai-apps must come from the same checkout: this app pins both to its own
# setuptools_scm version, which only exists on PyPI at a release tag.
python -m pip install -e . -e konfai-apps -e apps/impact_seg

⚙️ Usage

The CLI is organised into sub-commands, mirroring the KonfAI Apps operations:

Run anatomical segmentation on an input volume:

impact-seg-konfai segment body -i path/to/image.nii.gz -o ./Output/

Evaluate against a reference, or run everything at once:

impact-seg-konfai eval body -i image.nii.gz --gt reference_mask.nii.gz -o ./Output/
impact-seg-konfai pipeline body -i image.nii.gz --gt reference_mask.nii.gz --mask eval_mask.nii.gz --gpu 0 --tta 2 -uncertainty

Arguments

When --ensemble, --tta, and --mc are left at 0, the values declared in the app bundle (app.json) are used.

See the full help of any sub-command with:

impact-seg-konfai segment --help

📚 References

If you use IMPACT-Seg-KonfAI in your work, please cite KonfAI along with the IMPACT-Seg materials associated with the model release you use.

• Boussot, V., & Dillenseger, J.-L. (2025). KonfAI: A Modular and Fully Configurable Framework for Deep Learning in Medical Imaging. arXiv preprint arXiv:2508.09823

⚡ Performance & VRAM

Benchmarked on a single NVIDIA RTX PRO 5000 (24 GB) with a real whole-body CT (295 × 259 × 219, 2 mm), patch [1, 192, 192], single model. The app measures its batch size on your GPU (a forward of one patch, then of two, then the largest power of two that fits half of the free VRAM); override it in SlicerKonfAI (⚙ Advanced) or on the CLI with --patch-size / --batch-size.

Single-model body segmentation keeps system RAM ~1.6 GB. The thin 2-D patches never fill the card, so inference stays compute-bound (~7 s, largely batch-independent). Inference scales with the case size.

Measured on the S/M/L cases of benchmarks/perf/bench_apps.py (2026-09-09, body, default ensemble, same GPU). Peak RAM = process-tree resident set; peak VRAM = over baseline.

🔗 Links

• 🤗 Model Hub: huggingface.co/VBoussot/ImpactSeg
• 📦 PyPI Package: pypi.org/project/impact_seg_konfai
• 🧠 KonfAI Repository: github.com/fideus-labs/KonfAI