Project: Automatic labelling of Radiology Images

Summary: Create a tool to automatically label specified disease like pnumothorax, hemorrhage etc from xray images.

The tool should be developed as an standalone tool, and should have the options to draw labels of image segmentation and bounding box detection automatically from ML models. Automatic labeling would assist radiologist in their manual work.

Required Skills: Python, Deep Learning, Machine Learning, JS, ML Frameworks

Mentors: @pri2si17 @judywawira @sunbiz

NOTE

  • Do proper research of state of the art models for detection and segmentation, available opensource medical dataset and how a model can be created or enhanced.

  • Please develop a POC and provide link to your github repo. The basic POC should have the facility to upload dataset/images, choose labelling type (bounding boxes for POC), and the pretrained model should be able to label it. You can provide options to select model as well (like Faster RCNN, Retinanet, Yolo, etc). It should be a standalone web app, preferably in Python frameworks (Django, Flask) but other frameworks are also considered.

DELIVERABLES

  • Comprehensive research of existing opensource datasets, frameworks and different state of the art models. It should be google doc or pdf with link shared here.

  • Minimal web app to let user add/upload image and tool should be able to provide label for it according to the deep models selected by user.

  • Github repo should be shared here.

BONUS

  • Students are free to improvise the features and bring their own features with proper reasoning.

The requirement for main project will be updated over period of time, but POC should have the features mentioned above.

No use of Google Drive for sharing Proof of Concept (POC) applications.

3 Likes

I have relevant skills and want to contribute to this project , any prerequisite to follow?

Hi Vaibhav,

Welcome to the community. Its good to know that you have relevant skills but we would like to see some small POC developed for it. You can make a small web app and push your code on github. We would use that to see your skills. Also you can contribute to the radiology project. For data you can use opensource datasets from kaggle.

Reach out to us in case of any query.

Hello Priyanshu, I have worked with disease classification before. I did Diabetic Retinopathy using CNN, dataset-api and input pipelining(I have also contributed in tensorflow-datasets). I would like to know a little more specifics about the tool we should be making. Also for classification task, I believe a frontend react app with tensorflow.js would be a good Idea as it will put less burden on the servers. Do let me know if you like the idea, so that I can start working on it. Thank you.

I have already made a web app based on Django(python framework), as It is in private repository, if you share your Email/Usrename then I can share it with you.

Hi @SinghKislay , its good to know that you are familiar with this. I will add more specification to this project. Till then you can make small prototype for it and share the link. Feel free to reach us.

Hi @vaibhav, you can make public repo and share the link here. :slight_smile:

Hello @pri2si17 Priyanshu, I have previously worked on Diabetes and Sepsis classifiers models, both of which have pretty decent Accuracy. I want to contribute in this project as this matches my area of interest and I can deploy this as both web and Mobile App(Android).

Hi @sourabhyadav, welcome to the community. Please develop a POC and provide a github repo.

Thank you for updating “Note”

One thing to also note on what we’re looking for: We want to see that you can do research and work independently – GSoC is not about the mentor holding your hand the whole time.

3 Likes

Hi @pri2si17 I would like to contribute to the project. I have worked on fully-automated classification of OCT B-scans as a part of my bachelor’s thesis, the doi of my publication: 10.1007/978-3-030-34872-4_17 (Barnwal S , Das V, and Bora PK. Deep learning based fully automated decision making for intravitreal anti-VEGF therapy. In International Conference on Pattern Recognition and Machine Intelligence 2019. Springer, Cham.)

The entire work can be found in the following repository: https://github.com/simranbarnwal/deeplearning-for-antivegf-therapy

Kindly let me know how I can start contributing. Thanks! Projects I have undertaken so far: https://sites.google.com/view/simranbarnwal/projects_1

@pri2si17 could you do me a favor and add:

  • Deliverables that you want at a bare minimum for the student to complete
  • Stretch goals if applicable as well would be nice, but not necessary.

hello, @pri2si17 my name is Harsha. I have previously worked on computer-aided detection of interstitial lung diseases. I want to work on this project as this matches my area of interest.

Folks, please refrain from using Sir/madam. Thanks.

@pri2si17 @judywawira @sunbiz

Is that tool should be developed from scratch or we can use API

Hi @HarshaSatyavardhan,

Tool will be developed from scratch and in the process you can use different APIs needed. This will be later on integrated with toolkit or radiology. @judywawira and @sunbiz can give more insights over it , but it would be an standalone tool for now.

Hi @simranbarnwal,

Welcome to the community. You have experience working in this domain and you are welcome to contribute to this project. Please do as mentioned in the notes. Feel free to ping us in case of any doubt… :slight_smile:

@vaibhav, @SinghKislay @sourabhyadav

Any progress?

I was learning algorithms, currently on Mask-RNN, soon I will update you with Github repo. Data i am considering is https://nihcc.app.box.com/v/ChestXray-NIHCC containing data of 8 diseases.