Re: Inquiry Regarding U-Net in Biomedisa
Philipp Loesel
Hi Xiao,
I am pleased to hear that Biomedisa supports you in your work.
Biomedisa interpolation is an image processing or computer vision
task without the use of machine learning. Basically, statistical
information such as mean and variance is extrapolated from the
pre-segmented slices to the remaining volume. While it would be
possible to train a neural network on the pre-segmented slices and
apply it to the remaining slices, I haven't been able to get good
results when I tried that some time ago in 2020. This may have
changed somewhat with new developments and perhaps the use of
pre-trained networks. But due to my experience, the simplicity of
the Biomedisa interpolation and the fact that it works quite well
for many scenarios, I never used AI for the interpolation and
focused more on training networks for segmenting series of 3D
images.
Cheers,
Philipp
Dear Dr. Philipp Lösel,
My name is Xiao Fan from the University of Saskatchewan. Our team has been actively exploring biomedical image segmentation techniques, and Biomedisa has become one of our main tools. We have encountered a question on the interpolation part within Biomedisa. Specifically, we are curious about whether U-Net or any other deep learning techniques play a role in this process. While we understand that Biomedisa can train a CNN on image files along with corresponding fully segmented label files, we are keen to know if U-Net is also used in the interpolation function.
Thank you for your time and consideration.
Best regards,
Xiao Fan
_________________________________________________ Dr. Philipp Loesel Department of Materials Physics Research School of Physics (RSPhys) The Australian National University (ANU) 58 Mills Road, Cockcroft, Room C4.40 Acton ACT 2601 Australia email: philipp...@anu.edu.au https://physics.anu.edu.au/contact/people/profile.php?ID=3160 https://biomedisa.info/