Copyright and sources

• Angio Mapper, its network and its trained weights are © 2026 NeuroEvidence.ai. The network was designed and trained by NeuroEvidence.
• The trained model and the evaluation script are available to researchers for non-commercial research on request: contact us. They are not licensed for commercial use.

• Cerebral angiograms (training, checkpoint choice and testing). DIAS: Liu W et al. DIAS: a dataset and benchmark for intracranial artery segmentation in DSA sequences. Medical Image Analysis 2024. Distributed on Zenodo under CC BY 4.0. 30 frames were used for training, 10 to choose the checkpoint and 20 (sequences 40 to 59, the dataset’s own test split) for testing only.
• Sample images. The two samples on the Angio Mapper page are DIAS images. The yellow vessel map on them was added by NeuroEvidence.
• Pre-training. 1,626 X-ray angiograms with traced vessels from public research datasets.

• Test set. 20 hand-labelled angiograms (DIAS sequences 40 to 59), last-frame expert masks, 800 x 800 pixels, threshold 0.5, loose pieces under 25 pixels removed.
• VSS-Net. Single-frame result as published by Liu W et al., Medical Image Analysis 2024, on the same test images.
• clDice. Shit S et al. clDice: a novel topology-preserving loss function for tubular structure segmentation. CVPR 2021.
• Frangi filter. Frangi AF et al. Multiscale vessel enhancement filtering. MICCAI 1998. Run with scikit-image, threshold tuned on the 10 checkpoint-choice frames.

• ONNX Runtime (MIT licence) runs the network on our server.
• Tesseract.js (Apache 2.0 licence) reads burned-in text for the de-identification check.

Angio Mapper is an educational and research tool. It is not a medical device and is not for diagnosis or treatment. © 2026 NeuroEvidence.ai · Back to Angio Mapper