Load a brain scan to begin.
Load a brain scan to begin.
A browser implementation of SynthSR, developed by Iglesias, Billot and colleagues. Uses the pretrained v2 model and reproduces its 1 mm resampling, normalization, flip averaging and sharpening workflow.
Reference implementation · Iglesias et al., NeuroImage (2021)
Full-volume mode follows the reference method. Tiled mode reduces memory use but changes context and may introduce boundary differences. Outputs are synthetic images for research; they are not acquired T1 scans.
CPU inference uses ONNX Runtime Web 1.29.0; GPU inference uses optimized FP32 kernels; visualization uses NiiVue 1.0 RC. Source and model attribution: Apache-2.0.
Your images are processed locally in this browser and are never uploaded. The model weights download once from the Neurodesk asset store and may be cached on this device. The optional example images download from public repositories.
Download the native package for your system. The model is included, so these builds run offline without Python, Node.js or FreeSurfer.
Open the signed and notarized installer. Requires macOS 13.4 or later. Download for macOS · SHA-256
Extract the zip in PowerShell. No administrator access is needed. Download for Windows · SHA-256
Requires glibc 2.38+ and GLIBCXX_3.4.32. Keep the included shared library beside the executable. Download for Linux · SHA-256
Replace input.nii.gz with your scan. Add --ct for CT. Output includes the synthetic image and a JSON report. CPU is the default on Windows and Linux. Use --help for all options.
Download the Node.js package. Install it with Node.js 22 or later. The first run downloads and verifies the model.
npm install --prefix ./synthsr-cli ./neurodesk-synthsr-__PACKAGE_VERSION__.tgz
./synthsr-cli/node_modules/.bin/synthsr input.nii.gz output_synthsr.nii.gz --threads 8