Deep-learning On-chip DSLM Enabling High Volumetric Rate Three-dimensional Fluorescence Imaging And Its Application

Studying the neural activities in live organisms is important to neuroscience.However,volumetric imaging of dynamic signals in a large,moving,and light-scattering drosophila larva is extremely challenging,owing to the requirement on high spatiotemporal resolution and difficulty in obtaining signals with sufficient contrast.Through combing a microfluidics chipenabled digital scanning light-sheet illumination(DSLM)strategy with deep-learning based image restoration,this challenge could be overcome.Light-sheet microscopy(LSM)characterized with high spatiotemporal resolution has been widely used in fluorescence imaging.Combined with microfluidics which significantly increases the manipulation efficiency and throughput for LSM,and deep-learning technique that computationally improves the image quality,the performance of LSM can be further notably improved.The main content of this thesis includes:A DSLM setup on a conventional inverted microscope is designed and constructed,to enable 3D optical sectioning of freely moving drosophila larva ventral neural cord(VNC)at a volumetric rate up to 20.The 8-?m-thickness light-sheet covers a 500-?m-width field of view(FOV),and the spatial resolution is 2 ?m × 8 ?m.A three-layer microfluidics chip is designed and fabricated,to host live drosophila larva inside a chamber,thus allowing it freely moving within the FOV.The 3 mm × 0.5 mm chamber matches the FOV well.With optically-flatten side wall that led to smooth air-PDMS interface,horizontal plane light-sheet illumination notably improving the acquisition speed through the reduction of scanning range.Raw images are restored by deep-learning-based denoising and isotropic resolution enhancement.The intensity accuracy of reconstructed signals is over 92%,with an ?3 times increased signal-to-noise(SNR).The Drosophila locomotion patterns and corresponding neural activities are analyzed for forward and backward crawling larvae,revealing the corelations between neuron activities and worm locomotion.