💊Medical image analysis with Supervisely – Human in the loop AI

Share it with your friends Like

Thanks! Share it with your friends!


How to build blood vessel segmentation in retina images when you only have 6 images in training set.

Read the whole blog post on Hackernoon: https://hackernoon.com/deep-learning-in-medicine-advancing-medical-image-analysis-with-supervisely-33e936159206

More at https://supervise.ly


Dmitry Loktev says:

First step: uploading annotated images. What nature has annotated dataset in Supervisely (or elsewhere)?
Annotated data != training dataset. Annotated data is small in quantity, traing dataset is automatically generated from annotated data. This is what DTL is used for.

How to create a smart neural network, all process in steps:
1) annotate small amount of images manually (by human)
2) create more of them by using DTL (e.g. create a huge training dataset)
3) train an existing architecture on just ctreated training dataset (training architecture == creating your own model)
4) run your model on new images (they will be AI-annotated)
5) verify by human and correct errors (perfection it)


Write a comment