• Admin

Virtual Staining - Actin



This model transforms 2D brightfield images to fluorescence. The model simulates fluorescent stains targeting actin.


The training data for this model includes pairs of images captured at 20x magnification on differential interference contrast (DIC) microscopy for input and fluorescence microscopy for ground truth. The model is trained by Pixel2Pixel [1] with RCAUNet (U-Net with residual channel attention blocks) [2] as the generator and least square generative adversarial network (LSGAN) [3] as the discriminator.



Download


Requirements

Make sure you have Python versions 3.5 to 3.7 installed with TensorFlow-GPU version 1.13.0 and above.


Installation and apply instructions

Aivia is required for applying the model file. You can request a demo copy of Aivia here.

  1. Drag-and-drop the model file into the Recipe Console area; or use the 'Load recipe' option in the Recipe Console to load the model file.

  2. Load the test image (or any image of your own) into Aivia.

  3. If your image contains more than one channel, click on the 'Input & Output' section and specify the image channel you wish to apply the model on.

  4. Click 'Start' to apply the model.


References

  1. P Isola, et. al. Image-to-image translation wwith conditional adversarial networks. ArXiv.

  2. O Ronneberger, et. al. U-Net: convolutional networks for biomedical image segmentation. ArXiv.

  3. X Mao, et. al. Least squares generative adversarial networks. ArXiv.