Hi Ibrahim M. Ibrahim ,

I am completing my Ph.D. in the field of medical DeepLearning applications and am a MD by training. I am fully self-taught and it is doable. Just as you did, I started with Coursera and worked through DeepLearning.Ai's courses. Once you get behind the mathematics and are secure with Tensorflow 2 it isn't that hard.

My advice would be to just jump right into it. Most biomedical applications of DL are not that different from general problems and can be successfully tackled by the same approaches. Learning by doing...

Dear Ibrahim, deep learning is only a suggestive word for indicating multi-layers neural networks, I totally agree with Thomas: there is no specificity of biomedical problems with respect to other pattern recognition problems. The specificty is linked to the need of 'explainability' : we cannot accept 'pure black box' solutions in a field like biomedicine that present an high 'context dependence' and this point is of utmost importance. You must privilege the possibility to give a biologically sound explanation to the output with respect to the mere prediction accuracy, see:

Article Avoid Oversimplifications in Machine Learning: Going beyond ...

and https://www.sciencedirect.com/science/article/abs/pii/S1476927119302440

for some examples.

Alessandro Giuliani Tomas L Bothe Thank you for your advice and suggestions

Dear Ibrahim,

for starters, of course it can greatly improve your learning experience if you could find a supervisor because experienced people can always give you directions on what to learn next and which concepts are more important than others. But it is totally doable to learn it by yourself.

I think it is very good that you already took the initiative to learn the basic concepts of deep learning at coursera. Furthermore, I also recommend to split your learning in multiple directions. 1st direction is to learn the concepts from a theoretical point of view. 2nd direction is to code the algorithms yourself, e.g. coding a CNN, RNN, LSTM from scratch provides you with great insight into the different architectures and what kind of advantages/disadvantages they have. 3rd direction would be to make you familiar with the basic architectures out there, e.g. inception network, mobile net, efficient net, resnet, etc. in order to get a feeling what kind of architectures already exist and how they are used and why they are used in a specific context. This direction is easily tackled by reading research papers. That takes me to the 4th direction, read old and new papers, but read them carefully and analyse them, don't read 5 papers a week, take your time and read e.g. 1 paper. Try to implement what is done in the paper yourself, also great exercise. The old paper should give you the insight into the oldschool architectures and the new papers give you hint about where the future of deep learning is ongoing, new architectures, new methodologies, new trends and so on.

I hope I could help you with my little advices. But the most important thing is to have fun on your journey!