Deep Learning technique and Convolutional Neural networks structure are used in many applications and in divers fields. So, I want to get a conclusion of their advantages and disadvantages.
CNNs have been very successfully used for various biomedical image analysis problems. We use them for example in our lab (http://seunglab.org/) to automatically reconstruct thousands of neurons from terrabyte-sized volume electron microscopy datasets.
However, a general problem of these deep-learning algorithms is that they intrinsically require vast amounts of human annotated training data. Depending on the quality, data size and complexity this can mean hundreds or even thousands of working hours. In addition, each segmentation problem usually requires extensive hyper-parameter tuning, meaning that the parameters as well as the CNN-architecture need to be adjusted for every dataset. This is tedious, very time consuming and requires a lot of experience - that's why people sometimes also call that approach grad-student descent ;-)
To facilitate this kind of analysis, we founded the high-throughput image annotation service https://ariadne.ai/. Our generic workflow is quite simple: First we generate massive amounts of human annotated training data with our professional image annotators. Next, our machine-learning experts design a suitable image processing pipeline consisting of different classical image processing techniques and CNNs that are trained using our high quality training data. For example, in a large volume electron microscopy stack we automatically segmented almost half a million mitochondria as well as all the endoplasmatic reticula, both with very high accuracy (F1-Score of >0.99). Similar analysis pipelines also work for light microscopy data. We developed for example a CNN-based pipeline for the automated segmentation of all the somata in whole mouse brain light-sheet microscopy stacks with very high accuracy.
We would be thankful, if you can tell us more detailed structure of your CNN for one example of your published results and how long is the simulation time?
CNN is a kind of deep learning model and achieved promising results in image classification tasks. In other words, CNN acts as a powerful image classifier.
This link explains the usage of CNN as classifier for hand written digits.