Geometric Understanding of Supervised and Unsupervised Deep Learning for Biomedical Image Reconstruction

Explore a comprehensive lecture on the geometric understanding of deep learning in biomedical image reconstruction. Delve into the theoretical framework that explains why deep learning architectures outperform classical algorithms in inverse problems. Discover the unified approach that optimizes CNN design for various applications. Learn about a generalized cycleGAN framework for unsupervised learning in inverse problems without matched training data. Examine experimental results from supervised and unsupervised neural networks in biomedical imaging reconstruction to verify the geometric understanding of CNNs. Cover topics including deep learning for image reconstruction, diagnosis, and analysis, classical methods for inverse problems, input space partitioning, Lipschitz continuity, ultrasound acquisition modes, CT reconstruction approaches, and unsupervised learning for accelerated MRI.

Intro
Veep Learning for Image Reconstruction Diagnosis & analysis
Deep Learning Revolution for Inverse Problem
Classical Methods for Inverse Problems
Input Space Partitioning for Multiple Expressions
Lipschitz Continuity
Regularized Recon vs. Deep Recon
Ultrasound Acquisition Modes
Adaptive Beamformer
Image Domain Learning is Essential?
Two Approaches for CT Reconstruction
DBP Domain ROI Tomography
DBP Domain Conebeam Artifact Removal
Style Transfer : Power of Tight Frame U-net
Our Penalized LS Formulation
Unsupervised Blind Deconvolution Microscopy
Unsupervised Learning for Accelerated MRI
Summary