How do I edit this website?

NIA

DOI

NIA (Neural Inference Assistant) is a native, AI-powered denoising plugin that brings deep learning to ImageJ/Fiji without the configuration headache.

Developed by Dr. Kui Wang at the Center for Excellence in Brain Science and Intelligence Technology (CEBSIT), Chinese Academy of Sciences, NIA is built on the embedded ONNX Runtime. It allows biologists to run advanced denoising models directly in Java, eliminating the need for external Python environments, Conda, or complex CUDA setups.

Author: Kui Wang, CEBSIT, CAS. For questions please use the GitHub Issues or tag @Epivitae on image.sc.

Key Features

  • Zero Configuration: Runs natively in Fiji. No Python installation, no Conda environments, and no dedicated GPU setup required.
  • Proven Architecture: Comes with a built-in, optimized DnCNN model (based on Zhang et al., 2017) specifically tuned for fluorescence microscopy.
  • Extensible: Supports loading custom user-trained models in the standard .onnx format.
  • 5D Hyperstack Support: Seamlessly processes complex datasets (X, Y, Channel, Z-Slice, and Time-lapse) with auto-iterating logic.
  • Smart Normalization: Auto-detects bit-depth (8/16/32-bit) and applies consistent normalization to prevent “flickering” artifacts in time-lapse videos.

Installation & Update

NIA is distributed via the Biosensor Tools Fiji Update Site.

  1. Open Fiji / ImageJ.
  2. Navigate to Help › Update…
  3. Click Manage update sites.
  4. Check Biosensor Tools from the list.
  5. Click Apply and Close, then restart Fiji.

If the site is missing from the list, you can add it manually:
Name: Biosensor Tools
URL: https://sites.imagej.net/Biosensor-Tools/

Usage Guide

  1. Open Image: Load your noisy image or stack in Fiji.
  2. Launch: Go to Plugins › Biosensor Tools › NIA Denoise (AI).
  3. Select Model:
    • Built-in (DnCNN): Recommended for general fluorescence microscopy (confocal/widefield).
    • Custom ONNX: Select this to load your own .onnx model file.
  4. Run: Click Start Denoising. The plugin will process the stack and output a new denoised window.

Citation & References

If you use NIA in your research, please cite the software DOI:

Wang, K. (2026). NIA Denoise: User-Friendly AI Denoising Plugin for ImageJ/Fiji (v1.0.0). Zenodo. https://doi.org/10.5281/zenodo.18244343

Methodology Reference

The built-in model relies on the DnCNN architecture. Please also credit the original methodology:

Zhang, K., Zuo, W., Chen, Y., Meng, D., & Zhang, L. (2017). Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image Denoising. IEEE Transactions on Image Processing, 26(7), 3142–3155.

Developed by Kui Wang © 2026. Part of the Biosensor Tools suite.