[MISC] Ultrasound simulation model

[Rakshith2597]

Submitting training module for breast ultrasound simulation model.

This is part of the project MIRIAD: Many Incarnations of Screening of Radiology for High Throughput Disease Screening via Multiple Instance Reinforcement Learning with Adversarial Deep Neural Networks, sponsored by INTEL TECHNOLOGY INDIA PVT. LTD.

Principal Investigators:

Dr Debdoot Sheet (PI), Dr Nirmalya Ghosh (Co-PI) Department of Electrical Engineering Indian Institute of Technology Kharagpur

Dr Ramanathan Sethuraman (Co-PI) Intel Technology India Pvt. Ltd.

[Rakshith2597]

• Added a demo application that would allow other users to see what this project does.
• Export code for exporting Pytorch models to the ONNX and OpenVINO IR formats has been added.
• Minimal unit tests added.
• Details regarding metric used to evaluate the quality of model has been provided in ReadMe file.

[morkovka1337]

Run tests

[morkovka1337]

Please, fix pylint issues. You can further check this locally: pip install pylint and pylint --rcfile ../../../.pylintrc -v src && pylint --rcfile ../../../.pylintrc -v tests when you are in your project directory (breast_ultrasound_simulation). Also, there is no setup.py file that allows to install your training code as a python package via pip.

[morkovka1337]

Looks like you have accidently uploaded a lot of images instead of toy dataset...

[Rakshith2597]

Looks like you have accidentally uploaded a lot of images instead of a toy dataset...

Actually, it was not a mistake. Our model can simulate full 3D volumes using its multi-linear separable design by training the model only on 2D ultrasound images. To be precise, Given a CNN trained on a 2D ultrasound dataset acquired using a transducer of a particular frequency, we can simulate 2D ultrasound images and 3D volumes corresponding to various transducer frequencies without the need for availability of additional training data in 3D and at a different frequency. In the dummy/toy dataset provided, the folder 'real_images' contains 2D images and 'real_images_3d' contains images corresponding to a single 3D volume. If a user wishes to evaluate the performance of 3D simulation, they can use the sample/toy 3D volume provided. That was the reason for keeping those images (folders: real_images_3d and stage0_3d). Note that these images are not mandatory for the unit tests since the model is trained only on 2D images. All evaluations provided in the code( loss, performance metric, etc.) are done on 2D images. If needed, I can provide the files belonging to the 3D volume in google drive and mention it in the ReadMe.

[morkovka1337]

If needed, I can provide the files belonging to the 3D volume in google drive and mention it in the ReadMe.

I think, this would be better option. We store only minimal example of data in git, and data more than, like 10-20 images, is more preferable to store on the GDrive.

[Ilya-Krylov]

@Rakshith2597 will you update it?

[Rakshith2597]

will you update it?

@Ilya-Krylov Yes, I will be. The proposed model was trained using the IVUS dataset which does not comply with the licensing criteria of training_extension. To abide by the required license clauses, we have retrained the model with a new dataset. At the moment we are working on reorganising the code and adding necessary unit tests. We plan to update this as soon as necessary internal clearances are obtained.

[Rakshith2597]

@morkovka1337 The master and develop branch has changed significantly from the previous version. I see that this PR has been rebased to the misc branch.

1. Shall I proceed with this PR to merge with the misc branch?
2. There are a few more models which are part of the project that we had planned to submit here. Shall I create a PR on the misc branch for those as well?

[morkovka1337]

@morkovka1337 The master and develop branch has changed significantly from the previous version. I see that this PR has been rebased to the misc branch.

1. Shall I proceed with this PR to merge with the misc branch?

Yes, non-templated models are in the separate misc branch now.

2. There are a few more models which are part of the project that we had planned to submit here. Shall I create a PR on the misc branch for those as well?

Yes

[Rakshith2597]

@morkovka1337 Was this reviewed?

[morkovka1337]

Sorry, did not see ready for review mark. Will review in the nearest future.

[nervana-ff]

Can one of the admins verify this patch?

[ryanloney]

@goodsong81 can your team look into this?

[Rakshith2597]

@goodsong81 Apologies for the delay in closing this PR. Resolved issues and incorporated the suggestions from previous maintainers as well. I have fixed issues raised in codacy as well (all issues expect the ones raised for readme file.).

[Rakshith2597]

Test results on local server.

Pylint output