Medical ImageNEt - helping machine understand Medical imaging

fastai
1

Hey,
I am Amritpal. I am from Medicine undergrad student and have been using Fastai in medical imaging one year back. I really love this community and where it is going.

Theoretically speaking, Imagenet consists of Natural images, hence using Pre-trained models on Imagenet are generally helpful. But what if we had pre-trained model on medical images, where model understands a bit of all types of images - maybe that could be a better Pre-trained model to use for this specific task.
I have started making list of medical categories to cover. Would love to have contributions from this community. Also, if someone has previous experience in making good datasets, that could be super great.

2 Likes
2

You might want to check out Tencent’s MedicalNet pretrained models.

I don’t think they have posted what datasets the 23 datasets models are trained on though.

2 Likes
3

I Amritpal,

I am a spine physician and I have my own medical images (x-rays) of the lumbar, cervical and thoracic spine. Is there a large database of pretrained models for spinal x-rays in general? I just started with the fastai course

Thanks

1 Like
4

@devsoftware
Welcome to the community.
Im not sure if there exists a dataset in this particular domain. Why dont you go to kaggle.com and see if you can find the dataset that you’re looking for? Infact, I’ve seen multiple datasets on spines itself! Cheers!

1 Like
5

These are few links to get you started.

http://lit.fe.uni-lj.si/xVertSeg/database.php#40
https://www.kaggle.com/ekojsalim/spinenet-checkpoints


1 Like
6

Few other i am aware of -
http://lit.fe.uni-lj.si/xVertSeg/database.php#40


https://deep-spine.de/
7

Thanks @bwarner, This is a good place to start.

These models are nicely curated. But comprise of 3D image datasets only. They have 2D images in their wishlist though.
Do you have experience in dataset creation?