Home

Awesome

Dr.SAM

Vazgen Zohranyan, Vagner Navasardyan, Hayk Navasardyan, Jan Borggrefe, Shant Navasardyan

[Paper] [Dataset]

Dr. SAM design

Dr. SAM is a comprehensive multi-stage framework for vessel segmentation, diameter estimation, and anomaly analysis aiming to examine the peripheral vessels through angiography images. It was validated on the new benchmark dataset consisting of 500 X-ray angiography images with 1,500 vessel segments. The dataset was collected annotated by two expert radiologists. The dataset is publicly available and can be downloaded for research purposes.

Setup

The code requires python>=3.8.

  1. Clone the code repository:
git clone git@github.com:vazgenzohranyan/Dr.SAM.git
  1. For installing dependencies, run following command in your virtual environment.
pip install -r requirements.txt
  1. Download and install the SAM model checkpoint. The model checkpoint can be downloaded from here.

  2. Download dataset and extract it to data folder in the root of the project.

Usage

To run the algorithm on the dataset, run the following command in your virtual environment after setting up the project.

python scripts/run.py --data data/

For the full list of arguments, run:

python scripts/run.py --help

or check the source code.

Benchmark dataset

Dataset consists of carefully selected images from 500 angiographic examinations of the pelvic-iliac arteries, carried out between 2018 and 2024 at Bad Oeynhausen Hospital and JWK Klinikum Minden, within their radiology departments. The focus of these examinations was the abdominal aorta below the renal arteries and the pelvic arteries. Of these images, 450 have a resolution of 386x448 pixels, and 50 have a resolution of 819x950 pixels. The dataset includes 170 images featuring at least one stenosis and 64 images with at least one aneurism. The dataset archive is organized as follows:

metadata.json: This file contains information about bounding boxes and anomalies in the images. The file is structured as follows:

  {
    "image_id": 1, // Image id in the dataset
    "anomalies": { // Anomalies in the image with the format {anomaly: [(x, y)]}
      "stenosis": [[270, 351]], // Coordinates of the stenosis, if no stenosis, the field is empty []
      "aneurysm": [[229, 388], [253, 361]] // Coordinates of the aneurism, if no aneurism, the field is empty []
    },
    "bboxes": [[158, 11, 134, 192], 
               [13, 175, 187, 256],
               [193, 147, 159, 246]] // Bounding boxes of the vessels of interest with the format [x, y, width, height]
  }

License

The code is licensed under the Apache 2.0 license.
Our benchmark dataset is available for non-commercial, research purposes only

Citing Dr.SAM

If you use Dr.SAM in your research, please use the following BibTeX entry.

@article{...,
  Will be added
}