Generative Clearing for Deep Tissue Imaging
Abstract: Investigating cellular level tissue architecture requires the imaging of intact biological samples from 3D volumes with high-resolution imaging methods, including confocal microscopy. A significant challenge for the quantitative analysis of such volumetric data using computer vision is the degradation of image quality at increased depths due to light scattering, absorption and optical factors. We here introduce a generative cycle consistent adversarial network (Cycle-GAN) to mitigate these effects, which exploits the property that since the tissue is self-similar, the appearance of the shallow layers can serve as a proxy for that of depth degradation-free data. The network model obtains a bi-directional mapping between the shallow and deep layers so that the restored deep layers resemble the shallow ones. We demonstrate this approach's utility on a dataset of images obtained by confocal imaging of thick cardiac tissue sections from the mouse. Our experiments show that the restored deeper layers are qualitatively and quantitatively similar to the shallow ones, that the restored tissue images are amenable to geometric analysis and that in general such an approach outperforms other methods both qualitatively, as well as quantitatively.
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
Paper Type: validation/application paper
Primary Subject Area: Image Acquisition and Reconstruction
Secondary Subject Area: Application: Histopathology
Confidentiality And Author Instructions: I read the call for papers and author instructions. I acknowledge that exceeding the page limit and/or altering the latex template can result in desk rejection.
0 Replies
Loading