Questions about training the model

SillenceHu commented 1 year ago

Can you provide the code for training the model

LWHYC commented 1 year ago

Hi, SillenceHu

We are currently in the process of tidying up our training code and will publish it when it's ready. Meanwhile, our existing model is fully functional and capable of localizing any position in 3D CT scans, so you are welcome to use it as is. We appreciate your patience and understanding. Please feel free to reach out if you have further questions or need any assistance.

jumbojing commented 1 year ago

同求!

nanshanvv commented 1 year ago

Hi, SillenceHu

We are currently in the process of tidying up our training code and will publish it when it's ready. Meanwhile, our existing model is fully functional and capable of localizing any position in 3D CT scans, so you are welcome to use it as is. We appreciate your patience and understanding. Please feel free to reach out if you have further questions or need any assistance.

Looking forward to the release of the training code

TinySpins commented 1 year ago

Looking forward to your release of training code. Do you have any plans to try to train on MRI data as well?

LWHYC commented 1 year ago

Looking forward to your release of training code. Do you have any plans to try to train on MRI data as well? Hi,

Thank you for your interest in our work.

We're currently in the process of finalizing our article for submission. We plan to release the training code within the next month. However, we will be updating the inference code sooner to ensure faster localization and more accurate segmentation results.

As for MRI data, collecting it is indeed challenging. We're actively gathering data in an effort to train MedLSAM-V2, aiming for cross-modality arbitrary localization capabilities.

NisaHameed commented 11 months ago

Hi

Is there any chance that the work is adoptable to 3D microscopic images?

LWHYC commented 11 months ago

Hi

Is there any chance that the work is adoptable to 3D microscopic images?

Sorry for my lack of expertise in 3D microscopic images. Our work, MedLSAM, is successful on 3D medical data based on two assumptions:

Relative Distance Regression: The spatial structure distribution of different individuals' bodies is relatively similar, so they can be projected into the same coordinate system.
Multi Scale Similarity: Images with locally similar pixel distributions correspond to the same region in different individuals.

If your 3D microscopic images also meet either of the above assumptions, I believe it may be adoptable. However, further adjustments and experiments would be required to confirm its applicability.