Hi Muhammad,

I would say there is no standard criteria to choose a particular method. Usually, they are data-dependent. So, a specific technique may work better with a given dataset than other.

For your particular problem (i.e., identify and classify diseases) you will likely try to do this by using image patches, and maybe some other information. From my experience, Convolutional Neural Networks (CNN) do a pretty good job doing this.

Regarding the question of advantages/disadvantages, do you mean advantages/disadvantages of deep learning in general? or between different deep networks instead?

As mentioned by Jose, CNN is good for Image classification but data dependent.

You start with simple implementations (link below)

https://www.mathworks.com/matlabcentral/linkexchange/links/3785-convolutional-neural-nets-on-matlab-and-c++-as-mex-files-that-work-identically

There are also free implementations of the most successful Google Image classification DL networks. You'll obtain them as examples by installing Caffe (found on GitHub), which is a fairly approachable toolset for DL experiments anyways, including a python API.

The one thing I guess you need to bear in mind, though, is if you really have the data to feed such a capable machine enough to learn from. Google thinks in dozens of millions of images to differentiate a few handful of concepts reliably. Those are training set sizes a hospital will hardly have at hand, or can share freely (at least where I live).

So my five cents of wisdom: don't exaggerate your expectations.