Image Reconstruction from Projections: Radon Transform & CT

Added:

Iterative Reconstruction and Deep Learning Approaches

While the Radon transform and Filtered Backprojection (FBP) serve as the mathematical foundation for CT imaging, they have significant practical limitations. FBP is highly sensitive to noise and requires a large number of projections, which translates to higher radiation doses for patients. In contrast, Iterative Reconstruction (IR) techniques and modern Deep Learning models offer a powerful alternative. Instead of assuming a perfect mathematical transform, iterative methods formulate reconstruction as an optimization problem, incorporating statistical noise models and physical characteristics of the scanner. This allows for high-quality image reconstruction from sparse or noisy data, significantly reducing radiation exposure and mitigating artifacts that traditional analytical methods cannot handle.

Understanding of 1D and 2D Fourier Transforms and their representation in the frequency domain.
Basic knowledge of multivariable calculus, specifically line integrals and coordinate transformations between Cartesian and Polar systems.
Fundamentals of digital image processing, including concepts of 2D signal representation and spatial filtering.
The physical principles of X-ray attenuation (Beer-Lambert Law) and how projection measurements are acquired.
Advanced tomographic geometries, such as Fan-beam and Cone-beam (FDK algorithm) reconstruction used in modern 3D CT scanners.
Iterative and statistical reconstruction techniques (e.g., Algebraic Reconstruction Technique - ART, and Maximum Likelihood Expectation Maximization - MLEM) for low-dose imaging.
Identification and mitigation of CT imaging artifacts, such as beam hardening, metal artifacts, and ring artifacts.
Applications of reconstruction algorithms in other medical and industrial modalities, such as PET, SPECT, and MRI.
92.4K views0likes1:08:34@RichRadkeOriginal Release: 2015-04-13

This lecture explains how CT scanners reconstruct cross-sectional images of the body from X-ray projections using the Radon transform and filtered backprojection algorithm. The Radon transform mathematically represents projections as line integrals of the image intensity along lines parameterized by angle θ and perpendicular distance ρ. The Fourier-Slice theorem establishes that the 1D Fourier transform of a projection equals a slice through the 2D Fourier transform of the original image. Unfiltered backprojection causes blurring because it oversamples low frequencies at the center of the Fourier domain; filtered backprojection solves this by multiplying each projection's Fourier transform by a ramp filter (|ω|) before inverse transforming and summing over all angles, producing a sharp reconstruction.