Project: Low Powered Models for disease detection and classification for Radiology Images

Yeah… That’s why I mentioned emulator. The performance will not be same as hardware, but it will be roughly around it.

Hi Priyanshu, I am Pranath Reddy. I am a final year undergrad student at Birla Institute of Technology and Science (BITS) Pilani. I have worked extensively on the application of deep learning in Medical Image Processing and Automated Diagnostic systems, specifically detection and classification of heart valve disorders using unsupervised and supervised models such as Deep Convolutional autoencoders, CNN and LSTM. I have also deployed deep learning models trained using Keras to iOS by converting the checkpoints to coreML models. The domain of this project greatly coincides with my research experiences and interests, and I would like to contribute as well.

Hi Pranath, welcome to the community. Sure, you can definitely contribute to this.

1 Like

@pri2si17 Hello Priyanshu, I think we can use Tensorflow Lite for deploying model mobile applications. How to get to work for Proof Of Concept of this project?

Hi @Miftahunajat, at this point it would not be good to jump for deployment. I would like you to develop an application (running on android emulator), and model should be placed in that mobile platform itself. Application should not be complicated, keep it simple.

Hi Priyanshu, I have some queries regarding the proposal. Can you please specify the disease or part we are looking for the classification problem. Thank you!

For example being able to identify abnormalities like a broken bone or pneumonia on a chest xray, or maybe even a stroke or other abnormalities on a Head CT or MRI.

Hey guys, I’m a second year Computer Science student at the University of the West Indies. I’m interested in this project and currently working my POC.

1 Like

Hi @Shreya, well it can be broken into more finer details like single class classification or multi class classification for radiology images (if an xray belongs to multiple diseases). Also, it can be extended for detection part as well. Please let me know if you have further queries.

Hi @Aaron_Boodoo, welcome to the community, please provide link to github repo you are working on.

Dear @pri2si17,

I’m a student researcher working on neural network compression. I’m well adept with the different optimization models for reducing the size of neural networks so that they can be deployed on mobile devices.

In the POC, I find that there are two tasks that have been mentioned: creating an emulator and going through existing models.

I have gone extensively gone through the literature and have done extensive experiments on reducing the size of densenet without compromising it’s accuracy too much. So, I would like your suggestions on how should I proceed further in this???

Thanks and regards Ashraf

Hi @Ashraf, great. Well the POC is not about creating an emulator, it is to run models on emulator (because possibly you will not have real hardware). You can explore optimization algorithms like quantization, pruning etc. Implement it on any deep model and showcase it. For small demo you can have a simple android emulator or use raspberry from qemu (explore it if you don’t know), make a concise document of your research and finally submit a proposal.

@Shreya @shushantkmr2 @gauthampkrishnan @ajarihantjain54 @rsdel2007 @RitwickGhosh @shashank @Pranath @Miftahunajat and other students, you can also look at this comment.

1 Like

Hi @pri2si17 , I am Meghana , 3rd year CSE undergrad at SASTRA University, India. I use Deep Learning to build models for classification of EEG signals and I am also working on seizure prediction methods. I find this project very much interesting and I believe that I have the skillset to work on this project . Looking forward to working with everyone.

Hi @pri2si17, I am Nirmal Suthar, a second-year CSE undergrad at IIT Kanpur. My prime area of interest is applied DL/ML. I am interested in this project and also want to contribute to this community. Thank you :slight_smile:

Hi, I am Milind Thakur currently pursuing my BTech in Electronic and communication from IIIT Naya Raipur. Looking forward to contribute to this organization in GSOC 2020.

Hi @Meghana @milind @nirmal.ps welcome to the community. Please go through above project description and some useful comments and proceed accordingly. :slight_smile:

@pri2si17. Thanks for the reply. I have implemented pruning. I have run it on the standard desktop system. Haven’t yet tried on running it in an emulator. I have also started preparing a document regarding the other methods and techniques for reducing the size of the network that I have read. For the sake of the proposal is it alright if I do not run the model on an emulator for the time being???

Hey Priyanshu, I am Pratik Mahankal, currently pursuing Bachelors degree in Mehatronics from Symbiosis. I have strong background of machine learning and deep earning frameworks. And I have various projects in health care and radiology image processing. It would me pleasure to work for such a health organisation, working for such fascinating projects. Hope to work together. Regards, Pratik Mahankal

@Ashraf, can we see your code? A demo which will describe how much memory is consumed and what inference time you achieved before and after pruning.

Hi @pratikmahankal, welcome to the community. Please read project details (and comments) and provide us a git repository with your some awesome work. :slight_smile: