Project: Automatic Labeling of Radiology Images

This is the thread to be used for Project: Automatic labelling of Radiology Images

Hi @pri2si17, hope you are doing well. I wanted to ask, will we have the feature to login and email authentication in our web app?

Hi @Kislay_Singh, I think it is not on priority list. Can add later.

1 Like

Ok, I’ll just implement simple login for keeping track of the user and storing their inference results. We can do Email auths and JWT tokens later :slightly_smiling_face:

Hi @Kislay_Singh

Please fork the repository and use this docker for your further work. You can modify it according to your needs and raise a PR for it.

1 Like

Hi all! Under the guidance of @judywawira and @sunbiz, we finalized three radiology image labeling methods for integration to LibreHealth. Following is the list.

NYU Breast Cancer Screening System
This model is for detecting breast cancer in mammography images.

Code and pre-trained models are available on GitHub as mentioned in the paper.

Chest CT for PE

This is for PE detection in Chest CT images. Code and data can be requested from the researchers at Emory University. Please let me know when you need it. I will arrange the contact.

1 Like

Head CT

The following two papers have been selected for head CT processing. I will contact the authors for code and will update the forum on their response. We will decide on which one to use based on their response.

Please let me know if you have any questions!

1 Like

Hi @ATariq, I like the paper about Breast Cancer Screening System, there is a lot of data which will make our models perform great.

But, “Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning” has just 4,396 images, I’m afraid these are too low for such intricate deep learning task.

@Kislay_Singh The idea is to focus on the model that they are proposing. I have sent them a request for their code and we can ask for data as well. If they are not using public dataset, we may not get access to that. In that case, we will need to implement their model and train it over publicly available head CT datasets.

Similarly, breast cancer screening dataset may not be available but they have already made their code and pre-trained models available. We can use those.

1 Like

@ATariq, I went through the breast cancer paper and the repo mentioned there, I’ll get to it’s implementation in our project :slightly_smiling_face:.

[ Minutes of the meeting ]

  1. The app is for medical professionals to improve their efficiency of annotating data.
  2. The app should be having bounding boxes. (Any high performing object detection model)
  3. The app should have model versioning for ml models.

@judywawira @pri2si17, thank you for the input and time. If you feel like something is wrong or should be added please let me know.

Hi @pri2si17, in the light of our last conversation I have come up with a modified architecture for the project.

As the use case is very different from what I perceived initially, I have made changes such that our project will easily fit in any medical annotations pipeline easily with multiple annotators.

1 Like

This is just so that @Kislay_Singh can post again

1 Like

[ BLOG ]

Before Coding Begins: “Link

Hi @Kislay_Singh @aishwhereya @judywawira @geepriya @sunbiz

Lets do a discussion tomorrow at 9:30 PM IST (12:00 PM EDT). I know @judywawira will be busy, but rest please confirm your availability.


Hi @Kislay_Singh

Use this tool for documentation generation :

Let me know if you have any other queries.

1 Like

I will be available.


I am available too .

Hi @aishwhereya & @Kislay_Singh , It was nice meeting you both today. More information about this project requirement:

Excited to be working with you all this Summer. cc: @pri2si17


"Week 1"

Create the Django app that interacts with the user and tf-serving.

  • Write api for the users to interact.
  • Write api for the tf-serving to interact.

[User features]

  • Send bbox and receive Bbox info.(This includes which user did the annotations)
  • Flaging some areas of interest.(For later review or peer review)

[tf server features]

  • A dummy script that sends images with random bboxes to the django server.
  • Api that handles incoming images from the script.

"Week 2"

  • Training the tf model.
  • Setup tf-serving container.

"Week 3"

  • Complete the serving of tensorflow models into deployment.
  • implement the python middleware for facilitating communication between tf-serving and django.(I plan to make it like Facebook’s web hooks, the python middleware sends request to my django server.)

"Week 4"

  • Complete all backlogs, if remaining :slight_smile:
  • Complete the documentation.

Hi @pri2si17, this is my plan for the month of June.