Preprint: Introducing 3DRCAN for denoising, super resolution and expansion microscopy

Dear all,

We have recently completed a significant part of a colab project which characterizes and explores the limits of deep learning based solutions for image restoration (various scenarios, see below). We have put the results together in a pre-print which is now available on bioaRxiv. You are most welcome to test it out now.

The novel approach was tested on a range of organelles: Actin, ER, Golgi, Lysosome, Microtubule, Mitochondria. Moreover, it was tested using iSIM, Confocal / STED and Expansion Microscopy using both fixed and live cells (3D and 4D experiments). We also compared the 3DRCAN to other best in class neural nets, CARE, SRResNet and ESRGAN.

This work was done in collaboration with Hari Shroff, Jiji Chen et al. Most at the National Institute of Biomedical Imaging and Bioengineering.

More details (but still missing lots of cool stuff), in this twitter thread.

Aivia users will be able to use these models from Aivia 9.5 (planned for later this year)

All the best

Key details

Title: Three-dimensional residual channel attention networks denoise and sharpen fluorescence microscopy image volumes

Authors: J Chen, H Sasaki, H Lai, Y Su, J Liu, Y Wu, A Zhovmer, C Combs, I Rey-Suarez, H Chang, C Huang, X Li, M Guo, S Nizambad, A Upadhyaya, J Lee, L Lucas, H Shroff.


Open source (python) code and sample images:


Just a quick update. See (bottom of page) for a brief intro to and demo of some of the 3D RCAN models we have created for image restoration.