Related Experiment Videos
Artificial Second Harmonic Generation Image Feature Tuning with Generative Models and SeFa Semantic Analysis
Melissa Champer1, Vikas Singh2, Paul Campagnola1
1Department of Biomedical Engineering, University of Wisconsin-Madison, Madison, WI.
Jphys Photonics
|July 17, 2026
Summary
This study adapted StyleGAN2-ADA-SHG for analyzing collagen fiber morphology in Second Harmonic Generation (SHG) microscopy images. The model generates realistic images and identifies key fiber features, aiding disease-related morphological analysis.
Area of Science:
- Biomedical Imaging
- Computational Biology
- Materials Science
Background:
- Collagen fibers in the extracellular matrix change morphology with diseases like cancer.
- Second Harmonic Generation (SHG) microscopy visualizes fibrillar collagen morphology effectively.
- Deep learning methods show promise for analyzing microscopy data.
Purpose of the Study:
- Adapt StyleGAN2-ADA-SHG for SHG microscopy data.
- Extract further information content from SHG images.
- Develop tools for analyzing and editing collagen fiber morphology.
Main Methods:
- Trained StyleGAN2-ADA-SHG on 1,319 SHG images.
- Utilized Semantic Factorization (SeFa) to analyze generated images.
- Quantified fiber morphology using CT-Fire and CurveAlign.
- Tested encoder-based and LPIPS methods for latent space image editing.
Main Results:
- StyleGAN2-ADA-SHG generated novel SHG images perceptually similar to training data.
- Identified two semantics strongly correlated with collagen fiber morphology.
- The encoder-based editing approach outperformed LPIPS.
- Integrated analysis tools into a SeFa GUI.
Conclusions:
- StyleGAN2-ADA-SHG is suitable for generating and analyzing SHG collagen fiber images.
- Identified semantics can be used to create SHG images with variable morphology.
- Developed tools support future analysis and editing of SHG images for disease-related research.