Chest X-Ray AI

Anyone wanna help me work on this project to identify pathologies in thorax xrays using machine learning?
My Discord is ethan#8546
I have limited CS experience but know a lot of physiology/anatomy/med stuff. Contacted a leading researcher on this praying he gets back to me.

Attached: 00006585_011.png (1024x1024, 377.25K)

Other urls found in this thread:

tensorflow.org/tutorials/keras/classification
twitter.com/NSFWRedditImage

finally some good use of AI
bump

thanks

>page 4
damn, this thread is doomed
make a new thread some later, when there are more people on the board

replicate the results in "Automated abnormality classification of chest radiographs using deep convolutional neural networks" and find a new project
>but my use case is XYZ
replicate the results in the paper above, generate ~10k annotated images with the classification you want, re-train your model with your new annotated images.
>but replicating the results is hard/ generating 10k image annotations is hard
yes, or google would have already scooped you

Bump

From where do you plan to get your training data ?

Already have my training data

Attached: data.png (1012x686, 261.39K)

also
>clinical research
Can't do that until I have the thing built which is what i need help with

Retard, just see tensorflow docs. They have image classification examples of neural network.
Literally copy paste those ~20 lines and make your own neural network with out understanding anything

take a pre-trained model (eg AlexNet or VGGnet) with low error
you can find these off google
literally make only 1 change. replace the last layer with a sigmoid node that outputs probability of your disease.
retrain the model with your dataset after some parameter search
that's it, you should have a pretty strong classifier
this is from the methods section from the nature paper above

thanks

Agree with the other lads. Plenty of algorithms for abnormality detection already exist, you just have to hunt it down. If you have access to article databases from uni or work, just search up anatomy machine learning and you’ll find it.
The most basic thing you’ll find is registering sample images together to match up anatomical features, then use your test image and see how much abnormalities deviate from the reference.

I have no idea, but here take a bump for only interesting bread in the catalog.

It’s a breath of fresh air from all the “xxx linux is literally satan” and “cumfart programming language is for gypsies” posts

Your lymph nodes are swollen, might want to look into that

Subject has lung problems.

that dudes arm is broken, has bunch of ecg electrodes over his body and he looks old af. I think he's got bigger problems fr

This should be pretty simple to accomplish, desu. What you are trying to do is called "image classification" and is covered by some of the most basic Tensorflow tutorials: tensorflow.org/tutorials/keras/classification
The only challenge is compiling the data set. You basically need a ton of xrays and what amounts to a cheat sheet which can be used to train the AI.

>(((ethan)))
>work for free, goyim
every time

dude, it's consumer electronics board, here we discuss and argue about amd vs intel or shitpost about smartphones

You terrible gorilla nigger.

You should also use some healthy examples?