Details about Experiment on STARE Vessel Dataset

[THUeeY]

Thank you for your remarkable work. MedSAM2 used a mask prompt on the STARE dataset, as shown in Figure 4 of the paper. I would like to know whether the mask covers the entire vessel region in one large area, or if it is detailed enough to cover each vessel individually with an approximate mask. If you could guide me on how to perform testing and visualization on STARE, I would be extremely grateful.

[Songnansensetime]

Also want to know more details about how to achieve good results with vessel segmentation. What about 3D vessel segmentation? I wonder if it is possible to get comparable results.

[rabiaedayilmaz]

Hi,

I am still reading the paper, however, you can use perplexity.ai tool that can search internet and answer your questions with more-based answers. I gave paper link and your questions, and the answer:

In the context of the STARE dataset, the mask prompt used in MedSAM2 is designed to provide a detailed approximation of the vessel regions. According to the paper, the mask does not cover the entire vessel region in one large area; instead, it is crafted to delineate individual vessels more precisely. This allows for better segmentation performance, as the model can focus on the specific contours and characteristics of each vessel rather than treating them as a single entity.

Furthermore, as being new to concept, I searched through that tool and found out mmsegmentation toolbox. Seems it preprocesses data and probably used to prepare 2d dataset using: python tools/convert_datasets/refuge.py from mmsegmentation.

Here is the link to convertion: https://github.com/open-mmlab/mmsegmentation/blob/main/docs/en/user_guides/2_dataset_prepare.md

It has STARE dataset prepration guide.