CSBDeep in KNIME
This wiki page explains how you can setup KNIME to execute the example networks on different input data.
See the Installation Instructions for how to install and setup KNIME for the CSBDeep workflows.
- Each CARE network should be trained with data for a specific combination of image content (e.g., nuclei, microtubules) and image corruption (camera noise, pixel size, microscope PSF, etc.). Hence, applying trained networks to images that are very dissimilar to the training data could lead to unexpected results. The pretrained networks provided via Fiji and KNIME are meant to showcase our method on the accompanying example data.
These workflows reconstruct the data by denoising it in 3D. After that, a simple segmentation is done by thresholding the image. This gives only good results because of the clear reconstruction of the data.
See KNIME Workflow – 3D Denoising (Tribolium)
and KNIME Workflow – 3D Denoising (Planaria)
This workflow uses the surface projection network to find a fly wing in a 3D volume and map it onto a 2D plane.
See KNIME Workflow – Surface Projection (Flywing)
This workflow uses the isonet to improve the Z resolution of 3D microscopy images.
See KNIME Workflow – Isotropic Reconstruction (Retina)
This workflow uses a neural network to deconvolve an image of microtubules and make structures beyond the diffraction limit visible.