metadata
license: apache-2.0
library_name: konfai
pipeline_tag: image-to-image
tags:
- medical-imaging
- synthetic-ct
- image-synthesis
- radiotherapy
- mri
- cbct
- konfai
ImpactSynth — Synthetic CT (sCT) from MR / CBCT
Whole-body synthetic CT generation from MR or CBCT, built with KonfAI. Top-ranking models from the SynthRAD challenge (Tasks 1 & 2).
🧩 Models
| Model | Input | Output | Description | Ensemble |
|---|---|---|---|---|
MR |
MR | sCT | MR → sCT (SynthRAD Task 1) | 5 |
CBCT |
CBCT | sCT | CBCT → sCT (SynthRAD Task 2) | 5 |
MR_CBCT |
MR / CBCT | sCT | Joint MR/CBCT → sCT | 5 |
Finetune |
CBCT | sCT | BICMAC fine-tuned CBCT → sCT | 5 |
2.5D UNet++ · patch [1, 512, 512] · 5-model ensemble.
🚀 Usage
pip install impact_synth_konfai
impact-synth-konfai synthesize MR -i input_mr.nii.gz -o output/
- Generic runner:
konfai-apps infer VBoussot/ImpactSynth:MR -i input_mr.nii.gz -o output/ - Interactive: SlicerKonfAI — the ⚙ Advanced dialog overrides patch size and batch size.
⚡ Performance & VRAM
Benchmarked on a single NVIDIA RTX PRO 5000 (24 GB) with a real whole-body MR (295 × 259 × 219, 2 mm). The batch size is auto-selected from your free GPU VRAM.
| Free VRAM | Batch (auto) | Peak VRAM | Time / case |
|---|---|---|---|
| 8 GB | 16 | ~7.6 GB | — |
| 16 GB | 28 | ~15 GB | — |
| 24 GB | 32 | ~16 GB | ~24 s |
sCT generation keeps system RAM ~2 GB. The plan leaves memory headroom — filling the card with a larger batch saturates the allocator and slows inference. A full 5-model ensemble runs in ~82 s on the same card. Override with --patch-size / --batch-size.
🔗 Links & Citation
- 🧠 KonfAI: github.com/vboussot/KonfAI
- 📦 PyPI: impact_synth_konfai
- 📄 Paper: arXiv:2510.21358