Radiologist here. Not too worried about watson reading cat scans anytime soon. image segmentation is not a solved problem. Deep belief networks have been doing image classification, not segmentation. I welcome more competition in the radiology Pacs business though; which I think is all this is.
Thanks for the link. I myself have attempted using deep learning for segmentation with limited success with and without help from people with background in deep learning. I am not saying it's impossible but there is a lot of work left to do.
Yeah, with the renaissance that image analysis is undergoing in the last few years with DNNs, to say anything is impossible is showing a lack of imagination.
anyways, that's not my point. i'm saying deep learning for image analysis has been a huge success, and people should explore ways to apply this success to medical imaging.
but just for your info, neural networks are slowly being generalized to a lot of other challenges. look around for topics on recurrent neural networks, memory networks, reinforcement learning, etc etc. i don't think we've fully finished exploring the many ways neural networks can help solve life's challenges just yet.
Wow! Good. Thanks for this also. Last time I searched was a while ago, and I didn't find much. Glad to see progress. I will look into it "deeper" again.
Not everything is classification and there are problems better solved with other methods. For example I also haven't seeing anyone playing chess or go with DBN's. Also Compression, general programming...
Classification is a huge part of chess programs. You need to quickly evaluate a millions of boards to decide which side has the advantage. However, do to the well understood rules we can write efficient classifiers by hand.
Go is a much harder problem in large part because it's really hard to accurately classify which board is better off.
Teaching Deep Convolutional Neural Networks to Play Go: "Our convolutional neural networks can consistently defeat the well known Go program GNU Go... It is also able to win some games against state of the art Go playing program Fuego while using a fraction of the play time."http://arxiv.org/abs/1412.3409
You can say that about almost anything, and the world is still full of factory workers.
As a PhD student in medical imaging, you must also know that getting fully automating segmentation methods to work to the standard required in the clinic is really hard. And once you solve it for one clinic you will likely not be able to transfer the trained model to another clinic, because scan parameters, patients and workflow are different.
But when we solve the segmentation task, I think most radiologist will clap their hands and move on.
I follow the progress in deep learning too, Radiologists have been close to out of a job since the 80's, but this time may be different. Deep conv nets have basically allowed computers to surpass us in identifying visual patterns. The bottleneck is just the training data.
Are there applications of this technology in other areas of radiology as an automated second opinion? Would the culture of Radiology welcome such checks or would it be seen as a threat?
