Pneumonia has troubled the world for decades, with deaths mounting when early detection is missed. The World Health Organization estimates that each year the number of people affected by pneumonia is nearly 450 million, with the number of deaths reaching 4,000,000 per year, mostly in the developing world. These numbers represent a lot of pain, lost dreams and grief for people everywhere, and can be mitigated.
Hi @swapan. We are actually using x-rays for classification. There hasn’t been any development of a camera system that can penetrate beneath to scan human bodies. We are taking hard copies of x-ray images and then scanning them using the mentioned hardware which then classifies the image and lays down the results!
A few days ago I searched for public pneumonia x-ray images and found a repository at Kaggle. Created a EI-project to see if it’s possible to recognize if pneumonia is visible in the x-ray (of course it is). I’ll use this in a university ML-course I’m designing right now.
Today I thought to search this forum if this has been done before, and when I found this post, I remembered I’ve actually read it before! Some coincidence… but I’m glad I reinvented the wheel to learn a bit more
The accuracy I’m getting is 97% for int8 and 98% for float32. This after only 10 iterations with MobileNetV2 0.35, I’ve tried with more iterations, but EI is then timing out (not a problem in this case). A quick test by using my iPhone shows it’s also possible to use that to classify x-ray images as shown in the screenshot from my phone.
Surely…please let me know if you might face any issues in the deployment of the model!
You can even export the model as a WebAssembly file that you can run on any web-browser without connecting to EI Live Classification site everytime.