Why do you need to be trained on medical images?

If you don't want to train a neural network at all, then you need an architecture for a specific task. And if you plan to train on your data, then there is no big difference, the network is pre-trained on medical images or not. The first layers of the convolutional neural network are trained to identify low-level patterns that can be found in medical images.

I think. that applying ResNet with ImageNet weights, with the last layers re-learning on their data, would be a good start.

Hi Diana T Miranda ,

There are a number of ImageNet large scale databases for visual recognition, etc.

Check this one and see if you can download the data:

http://image-net.org/challenges/LSVRC/2016/index

http://image-net.org/challenges/ilsvrc+coco2016

Also check the following papers to see if you may contact the authors or access the data:

https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks

Article Very Deep Convolutional Networks for Large-Scale Image Recognition

Article Toolflows for Mapping Convolutional Neural Networks on FPGAs...

Thank you Mark Panenko and Bahman Sadeghi for taking time to answer my query.

I have already worked with CNNs which are pretrained on ImageNet. I have performed fine-tuning as well as retrained the entire CNN using medical images.

I now want to compare those results with those obtained using a CNN which is trained on medical images. However, currently I do not have sufficient resources to do this. Hence, I need a CNN which is pre-trained on medical images.

Short answer, there is not, and even if you find one, on which medical imaging modality do you want it to be pre-trained (e.g., MRI, X-ray, ultrasound, SPECT, PET)?

I want it to be trained on multiple modalities. I'm working on handling images of various modalities. So I wanted the CNN to be able to handle all of them or atleast majority of them.