Скачать или смотреть Linjing Fang - Deep learning-based point scanning super resolution imaging - Imaging ONEWORLD

“Deep learning-based point scanning super resolution imaging"

Point scanning imaging systems are perhaps the most widely used tools for high resolution cellular and tissue imaging. These systems benefit from the unique ability to arbitrarily set the pixel resolution of an image during acquisition. However, the optimal pixel resolution, speed, and signal-to-noise ratio (SNR) of point scanning systems are in tension with one another. Here we introduce a novel framework for restoring low SNR, low pixel resolution images to high SNR, high resolution images, which we term point-scanning super-resolution (PSSR). To address the limitations and costs associated with generating training data, we developed a ‘crappifier’ that generates semi-synthetic training data from pre-existing high-resolution datasets. Remarkably, our models could restore undersampled images acquired with different optics, detectors, samples, or sample preparation methods. For high spatiotemporal live cell imaging of mitochondrial dynamics, we developed a semi-synthetic multiframe training approach that facilitates otherwise impossible results with a normal point-scanning confocal microscope. In conclusion, PSSR facilitates point-scanning image acquisition with otherwise unattainable resolution, speed, and sensitivity.

Linjing is the Image Analysis Specialist in the Waitt Advanced Biophotonics Core at the Salk Institute. During her masters at Cornell University, Linjing was interested in understanding why brain blood flow is decreased in Alzheimer’s patients by analyzing deep brain capillaries imaged via 3D multi-photon microscopy. To speed up the pipeline for better interpretation of the vascular topology, she systematically tested several state-of-the-art algorithms for vessel segmentation, and determined that machine-learning based methods were superior to all other automated methods, greatly increasing the efficiency of the otherwise laborious task of manually segmenting images. In her current role as Image Analysis Specialist, Linjing works closely with researchers to help with quantitative image analysis, covering a wide variety of biological samples and image types including fluorescence widefield, Airyscan, confocal, STORM/PALM, 2-photon, transmission, scanning, and serial blockface scanning electron microscope images. She also develops custom algorithms for collaborators, and provides training to all users on image analysis software such as Imaris, Fiji, Arivis, and Aivia. For her independent project, she developed deep learning-based super-resolution image restoration software that is able to push the limits of resolution, photon/electron dose and imaging speed for fluorescence and electron microscope images with extremely low signal-to-noise ratios. She is also working on developing a deep learning-based algorithm for volume reconstruction from series of ultra-thin electron microscopy data with missing or damaged sections and software for high-throughput complex plant root segmentation.