Implement intervertebral disc detection using pose estimation

[jcohenadad]

After implementing the first deep learning approach for detecting discs spinalcordtoolbox/spinalcordtoolbox#2679, we could investigate the post estimation method: mathis2018.pdf

[rezazad68]

Just a quick update, the implementation of the pose estimation approach ( stacked hourglass network) for the training section is completed (not finalized). sample of validation result is attached. I Will update here regularly (in case of an issue) for further improvement and evaluation.
It seems to me the gaussian kernel is small and I need to fix it along with other things ...

[jcohenadad]

Very cool! Thank you for the report @rezazad68 . I would suggest another choice of color-- it is very difficult to see the blue labels.

[rezazad68]

I have created a temporary repository to host the implementation code. Please check this link and guide me to solve issues (check the Issues to handle section). I highly appreciate your comments. Issues mostly related to the dataset.

[jcohenadad]

I have created a temporary repository to host the implementation code. Please check this link and guide me to solve issues (check the Issues to handle section). I highly appreciate your comments. Issues mostly related to the dataset.

@rezazad68 I've briefly looked at the repos. It is very well documented, however I see a lot of code/feature duplication from ivadomed. E.g. the GIF generation and BIDS data loading are already implemented in ivadomed, so why not using them? That way, you can also create needed features in ivadomed, and it ensures your codebase will be compatible with ivadomed once your study is ready for publications. Currently it is not, there is a lot of hard-coded path/filenames in your code, eg:

• https://github.com/rezazad68/DeepSpine/blob/8b7d9347ae5fd04cbfe61536918f401230f18db2/src/Data2array.py#L22
• https://github.com/rezazad68/DeepSpine/blob/8b7d9347ae5fd04cbfe61536918f401230f18db2/src/Data2array.py#L158

[rezazad68]

Dataset Issue solved, as @jcohenadad suggested it was related to the wrong modality (T2w instead of T1w).

[jcohenadad]

Dataset Issue solved, as @jcohenadad suggested it was related to the wrong modality (T2w instead of T1w).

cool! in the future, when you see an issue, could you please open an issue directly in your repos (if the issue is specific to your project) or in ivadomed repos (if issue is specific to ivadomed)-- it helps centralizing discussions-- otherwise we end up with a thread of 500+ post which is difficult to follow-- thx

[ssinad]

GitHub supports transferring issues in between repositories of the same user or organization.
https://docs.github.com/en/github/managing-your-work-on-github/transferring-an-issue-to-another-repository

[jcohenadad]

thank you @ssinad in this case unfortunately this is another org