Technologist Track Session 1: CMR Technical Methods (2019)-Compressed Sensing

Video Transcription

Video Summary

The presentation explains compressed sensing for MRI, showing how it can reconstruct good images from heavily undersampled data. MRI is usually slow because it collects k-space line by line, so the goal is to acquire fewer samples while preserving image quality. Compressed sensing relies on three key ingredients: sparsity, incoherent sampling, and nonlinear reconstruction. Images are sparse or compressible in transforms like wavelets or finite differences, allowing noise and artifacts to be removed through thresholding, especially soft thresholding linked to L1 minimization. Undersampling should be random or variable-density so aliasing appears noise-like rather than structured, making it easier to separate from true signal. Reconstruction is done through iterative optimization combining data consistency with sparsity enforcement. The speaker also shows how the math relates to the MRI signal equation and mentions available software tools for implementation.

Keywords

compressed sensing

MRI reconstruction

undersampled k-space

sparsity

nonlinear optimization