How to go about manually segmenting images?

I have some unlabeled medical videos where the object is easy to outline by hand. Has anyone had experience manually annotating binary segmentation masks? How much data can I expect to label for some FCN like U-net to start being accurate?

Also, does anyone know any open source programs for manually annotating medical data?

1 Like

There is a program called sloth that is supposed to do this. I have used it in the past and its quite easy to use.
http://sloth.readthedocs.io/en/latest/

ITK SNAP is good for semi automated segmentation of 3D data - could probably use on videos too.

I’ll check out ITK SNAP. Seems to do exacting what I need. Thanks!

Hey Charles,

There are a few opensource options for segmenting medical data: ITK-Snap, DICOMworks, RIL-contour or 3D Slicer.

Labeling medical images is a pain! Especially for DICOM and NiFti imaging, we usually outsource outsource the job to workforce of doctors to annotate for us. It saves us a lot of time and energy.

Also, do try: https://annomed.io/

We are working to connect practitioners in Thailand with AI product developers worldwide to provide gold-standard annotations. You will get to see practitioner profiles and manage your project with breeze. We would love to hear your feedback.