I want to create chronic respiratory diseases detection using deep learning. i have basic knowledge of neural network , padding, tensorflow

i want to create this project for a hackathon, please guide me since we are only two person team

In the beginning you should take some courses to learn deep learning and we here have some pretty good ones, check them out!

1 Like

Hi Sahil. You mentioned that your objective is detection i.e. predicting bounding boxes. Therefore, I suggest that you start by studying this simple notebook that shows how to use Yolov5 to detect balloons:

This dataset contains Tuberculosis (TB) x-rays with bounding boxes. Using the workflow described in the above notebook, you can replace the balloon data with this TB data. You will then be able to detect TB.

This is a list of freely available TB and Pneumonia datasets that you might also find helpful.

I also suggest that you search for competitions and datasets on Kaggle that are related to your project and study the Kaggle notebooks that have been published. Good luck.


i did some deep learning courses like TensorFlow and convolutional yt course, thats why i have basic understanding, but if there’s any other course which is worth studying for this project do suggest me.

thanks dude for this.

I am not suggesting that using object detection - localization plus classification - on diagnostic imaging is a bad idea. But if you don’t know, imaging is not the only, and sometimes not the primary, diagnostic method for all chronic respiratory disease. Spirometry and blood gas analyses are two examples.

Also, if this is an exercise for refining your newly acquired deep learning chops that is fine. But it doesn’t seem to be breaking new ground, as there is considerable prior art. Probably worth a quick look to see what lessons or pitfalls you can learn from them…


Development and validation of an abnormality-derived deep-learning diagnostic system for major respiratory diseases


I suggest this often though no one ever takes me up on it, but you could add value to the community if you come back later and share your experience. What architecture did you end up choosing? How did it work? What were some strengths, weaknesses, lessons learned etc