KEYWORDS: 3D modeling, Second harmonic generation, Collagen, Structured optical fibers, Modeling, Tissues, Image processing, Diseases and disorders, Transmission electron microscopy, Principal component analysis
We have shown that diseases including cancers and fibroses have significant changes in the collagen fibril and fiber structure. We have used machine learning methods to classify normal and diseased tissues based on the fiber morphology in SHG images, however the important features remain unknown. Using the StyleGAN framework, we trained the latent space and used PCA for this determination. We found that curvature and density were the most important attributes in ovarian cancer. We previously demonstrated qualitative differences in the fibril size and spacing in several tissues. We now developed a more complete computational model of 3D phasematching. This in conjunction with wavelength- dependent measurements of the spatial emission pattern, will afford the extraction of average fibril size and packing. Collectively these determinations will be broadly applicable to identifying characteristic diagnostic targets in a wide range of diseases.
We have developed SHG microscope tools to probe all levels of collagen architecture organization in human high grade serous ovarian cancer (HSOC). We have found pronounced differences using machine learning classification of the fiber morphology as well as alterations in macro/supramolecular structural aspects through polarization analysis. We have used multiphoton excited fabrication to create SHG image-based orthogonal models that represent both the collagen morphology and stiffness of normal ovarian stroma and HGSOC. We found the fiber morphology of HGSOC promotes motility through a contact guidance mechanism and that stiffer matrix further promotes these same processes through a mechanosensitive mechanism. We have also developed a machine learning approach using generative adversarial networks (GANs) to optimize the scaffold design. Collectively, this data provides insight into disease etiology and suggests future diagnostic approaches.
Biological imaging studies are often limited by a low amount of data, decreasing the reliability of typical machine learning methods. We attempted to address this by creating large numbers of synthetic second harmonic generation images that can be tuned to reflect properties of different disease classes. To start, we collected collagen images from a variety of healthy and diseased specimens. These were analyzed with a modified generative adversarial network (StyleGAN) combined with an encoder. After training, we were able to produce images that accurately reflected our samples. These results can be applied to increase the accuracy of classification algorithms and models of extracellular matrix tissue.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.