Skip to content
Other

DSA vessel segmentation

in-browser vessel segmentation

2 min readupdated

A model for binary brain vessel segmentation from digital subtraction angiography (DSA) images. It consists of five folds (ensemble parts) of 2D nnU-Net v2 trained with the clDice loss function and exported to ONNX format without weight modification. The key feature of the model is its ability to run directly in a web browser using ONNX Runtime Web without sending data to a server. This development is intended for research purposes and is not a medical device.

Key features

  • Precise brain vessel segmentation based on the nnU-Net v2 architecture.
  • Full browser compatibility via ONNX Runtime Web with WebGPU support.
  • Five-fold ensemble for improved result accuracy.
  • Support for optional test-time mirror augmentation.

What you can do

  • Test the model locally in your browser through a demo application.
  • Integrate ONNX models into web services for client-side medical image processing.
  • Set up preprocessing and post-processing pipelines (Z-score, sliding window, Softmax) in Python or JavaScript.

Example

A minimal example of checking a single patch in Python using onnxruntime:

import numpy as np, onnxruntime as ort
s = ort.InferenceSession("vessel_seg_fold0.onnx")
x = np.zeros((1, 1, 896, 768), np.float32)          # z-scored патч
logits = s.run(None, {"input": x})[0]               # (1, 2, 896, 768)
vessel = logits[0, 1] > logits[0, 0]
Read moreDetails
Framework: nnU-Net v2 (2d configuration, PlainConvUNet, 46.3 million parameters per fold).
File format: vessel_seg_fold{0..4}.onnx (fp32, about 185 MB each).
Performance: processing a single fold on WebGPU takes about 1.1 s, and the full 5-fold ensemble takes about 4.1 s.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.2 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.0 GB0.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # cjy666/dsa-vessel-seg
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download cjy666/dsa-vessel-seg --local-dir ~/models/dsa-vessel-seg
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to dsa-vessel-seg
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of dsa-vessel-seg (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

cjy666/dsa-vessel-seg
Author
cjy666
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments