Explore advanced computational microscopy techniques in this IEEE Signal Processing Society webinar featuring Prof. Lei Tian from Boston University. Delve into LED array microscopy for multimodal computational imaging, focusing on intensity diffraction tomography (IDT) with controllable LED illumination. Learn about linear single-scattering approximation models, 3D reconstruction methods, and techniques to expand IDT's imaging limits. Discover multiplexed IDT for high volume-rate imaging, multi-scale phase tomography, and real-time volumetric phase imaging. Examine the limitations of single-scattering models and explore multiple-scattering approaches to improve reconstruction accuracy. Investigate deep learning strategies that leverage single-scattering inversion and learn multiple-scattering physics, enhancing weak scattering recovery and generalizing across different optical setups. Gain insights into lightweight 2D U-Net architectures for efficient 3D recovery, model-based data normalization techniques, and the use of multiple-scattering physical simulators for generating training data. Understand how neural networks trained on simulated data can generalize to biological samples, advancing the field of computational 3D phase microscopy.
Computational Microscopy Using LED Array for 3D Phase Imaging - Prof. Lei Tian
IEEE Signal Processing Society via YouTube
Overview
Syllabus
Computational microscopy using an LED array
LED array microscope for multimodal computational imaging
Intensity diffraction tomography with controllable LED illumination
Linear single-scattering approximation model Sample
Single scattering information is directly visible in the Fourier space
3D reconstruction by slice-wise deconvolution
Scan-free 3D phase reconstruction on cell cluster
Expanding IDT's imaging limits
Multiplexed Intensity Diffraction Tomography (MIDT) for high volume-rate IDT
Multi-scale phase tomography
Annual IDT (alDT) further extends imaging limits
Real-time volumetric phase imaging
"Open" computational 3D phase microscopy
Single-scattering model limits IDT's reconstructio
Multiple-scattering model to improve accuracy
Nonlinear SSNP multiple-scattering model
Reconstruction by optimization
Reconstruction on multiple-scattering sample
Reconstruction on dynamic sample
Deep learning training leverages single-scattering! inversion and learns multiple-scattering physics
Deep learning training leverages single-scattering inversion and learns multiple-scattering physics
Network improves weak scattering recovery
Network generalizes to different optical setups
Network generalizes to dynamic multiple-scattering sami
Acknowledgements
Model and learning strategies for Computational 3D phase microscopy
Lightweight 2D U-Net for efficient 3D Recovery
Multiple-scattering model-based reconstruction is still limited
Compute SSNP forward model stepwise
Model-based data normalization for generalization across different contrasts
Network prediction validation
Multiple-scattering physical simulator for generating training data Neural network trained entirely on simulated data generated using multiple scattering models
Trained deep learning model generalizes to biological samples
Taught by
IEEE Signal Processing Society
