Fig. 4: Improving image quality and throughput.
From: Neural network-based processing and reconstruction of compromised biophotonic image data
a Schematic representation of a neural network-enabled pipeline for image quality improvement of data taken with simplified and/or inexpensive optics. b Deep learning enhanced mobile-phone microscopy with a CNN trained to denoise, color-correct, and extend the depth of field with examples of blood smears and lung tissue sections. c Low exposure STED SNR enhancement through UNet-RCAN. The example shown compares noisy images (exposure time of 50 ns), ground-truth images (exposure time of 1 μs), and images processed by UNet-RCAN on β-tubulin (STAR635P) in U2OS cells21 [figure adapted from ref #21, licensed under CC BY 4.0, http://creativecommons.org/licenses/by/4.0/]. d Reconstruction of low power (LP) SRS coronal mouse brain images and deep learning denoised versions, as well as two-color (lipids-green, proteins-blue) SRS images of a coronal mouse-brain slice with the ground truth as high power (HP) SRS images19 [figure adapted with permission from ref # 19 © Optical Society of America]. e The process of overlapped microscopy imaging involves illuminating various independent FOVs of samples using LEDs, followed by capturing these through a multi-lens array onto a shared sensor, resulting in an overlapped composite image. A CNN-based analysis framework is applied to detect and identify specific features within this composite image. This technique is exemplified by the model finding a target from an overlap of 3 images17 [figure adapted with permission from ref # 17 © Optical Society of America]. f Example of the low light SIM pipeline. For training the U-Net model, either fifteen (using three different illumination angles (Nθ = 3) and five phase patterns (Nψ = 5)) with faint illumination or three SIM raw data images (a single phase pattern for fewer raw data acquisitions) are employed as input, while high SNR SIM reconstructions serve as the ground truth. This approach is shown with examples on microtubules18 [figure adapted from ref #18, licensed under CC BY 4.0, http://creativecommons.org/licenses/by/4.0/]
