[Solved] unusual models and algorithms that I could use to diagnose medical images?
So basically I have 2400. (64x64) images of lung X-rays. 1200 of those images are images of lung x-rays with pneumonia and the other half are images of lung x-rays that are normal. I've used 20% of the images for my training data set and the rest as my test data set. So far I've used the following models:
KNN("k" nearest neighbours), Logistic Regression model, MLP (multi layer perceptron), CNN (convolutional neural network) and VGG 16(transfer learning).
I noticed that the validation in the CNN model fluctuated a LOT. So the prediction accuracy of the final epoch was low but there was another epoch in the middle where there was a decent prediction accuracy (like 87%). I also changed the learning rate of the Adam optimisation tool and that slightly augmented my results.
So anyone here know of some other cool models that could perhaps be useful for my project? I was thinking about setting up GANS because I've anyway scaled my images down and GANS work well on small images but I'm not sure how I'd exactly implement it. Since I'm trying to determine wether or not an X-ray has pneumonia or not I think I'd have to have 4 models. One generator for pneumonia, one discriminator for pneumonia and then another generator for normal and then one more discriminator for normal. Even if I setup those 4 models, how would I eventually tie it all together? Please help.
So if you know about any other interesting or unusual models or you could maybe help me setup my GANs please let me know, it'll be super helpful. If you want you can add my discord - ExtendoClip#6259
you are using 20% of your dataset for training (480 images in total), you should increase the number of training data, for this sake you can apply data augmentation for the training set.
You can also find a similar pre-trained model with the same topic at Github, so you can also apply Transfer Learning to build your own model based on the pre-trained one.