Department of Biomedical Medical

Engineering Riphah College of Science and

Technology Faculty of Engineering & Applied
Sciences Riphah International University

Program: B.Sc. Bio-Medical Engineering Semester: VII

Subject: MedicalImaging BML-411 Date:

Experiment: 11
Calculate the inverse radon transform and compare filtered and
unfiltered back projection

Objective:
 To assess how well the filtered back projection reconstructs the original CT image compared
to unfiltered back projection, focusing on clarity, detail preservation, and noise suppression.
 To evaluate the impact of filtering on noise and artifacts in the reconstructed image, such as streaking or
blurring, and determine the trade-offs between smoothness and sharpness in the two approaches.
 To perform a quantitative comparison using metrics such as mean squared error (MSE), structural
similarity index (SSIM), or peak signal-to-noise ratio (PSNR) between the reconstructed images and
the ground truth.

Name: Sap ID:

Lab Performance:
Sr.No Title Marks Obtained Marks
1 Ability to conduct experiment 5
2 Subject Knowledge and Data Analysis 5
3 Total 10

Lab Report:
Sr.No Title Marks Obtained Marks
1 Organization and structure 5
2 Calculation and data presentation 5
3 Total 10

Remarks (if any)………………………………………………..

Name & Signature of faculty: …………………………………

Lab 11: Calculate the inverse radon transform and compare filtered and unfiltered back projection

Introduction
The Radon Transform is a mathematical operation that calculates the integral of a function along rays. It
is commonly used in tomographic systems, such as X-ray imaging in medical applications, to
reconstruct two- or three-dimensional objects from these integrals. The Radon transform is the transform
of our n-dimensional volume to a complete set of (n-1)-dimensional line integrals. The inverse Radon
transform is the transform from our complete (n-1)-dimensional line integrals back to the original image.

The Radon transform takes a 2D image and transforms it into a set of 1D line integrals at various
angles, creating a sinogram. The inverse Radon transform reconstructs the original image from these
projections, using back projection to map the line integrals back to the 2D image space. Filtered back
projection applies a filter to improve the reconstruction quality by sharpening the image and removing
artifacts.

Explanation:

Transforms

Both the Radon and inverse Radon transforms are fundamental to techniques like CT scans and MRI,
where images are reconstructed from a set of projections taken at different angles. The Radon transform
and its inverse play a crucial role in imaging techniques, particularly in medical imaging like
tomography and MRI, where they help reconstruct images from projections. Here's a more detailed
explanation:

Radon Transform:

The Radon transform of a 2D image (like a cross-sectional MRI image) involves taking line integrals of
the image. Mathematically, it can be expressed as the integral of the image intensity over a line, for
different orientations and positions of the line. This produces a sinogram, a 2D array that represents all
the possible projections of the image at various angles.

In 2D, the Radon transform is the mapping of the original 2D image to a set of 1D line integrals,
collected at various angles. Each line integral represents the sum of pixel values along a straight line
passing through the image at a specific angle.
The output is a 2D matrix (sinogram) where each row corresponds to a projection of the image at a
particular angle, and the columns correspond to positions along the line of projection.

Inverse Radon Transform:

The inverse Radon transform is the process of reconstructing the original image from its set of line
integrals (projections). Given a complete set of projections at all angles, the inverse Radon transform
can be used to recover the original 2D image.

Mathematically, the inverse Radon transform can be computed using a back projection technique, where
each projection is smeared or backprojected into the image space. The reconstructed image is essentially
the sum of these back-projected projections. However, back projection without filtering can result in
blurry images, which is why filtering techniques are often applied before the back projection to improve
image quality.

In general, the inverse Radon transform can be expressed as:

Filtered Back projection:

In practice, filtered back projection is used to enhance the inverse Radon transform. A filter (such as the
Ram-Lak filter) is applied to the projections before back projecting them. This process reduces blurring
and improves the sharpness of the reconstructed image. Without filtering, back projection alone often
results in images with blurred edges due to the interpolation of projection data from various angles.

MATLAB Code:

Code: 1

% Read the brain MRI image

img = imread('Brain MRI.jpg');

% Convert to grayscale if it's RGB

if size(img, 3) == 3

img = rgb2gray(img);

end
% Resize the image to a manageable size for computation

img = imresize(img, [256, 256]);

% Display the original image

figure;

imshow(img);

title('Original Image');

% Perform the Radon transform

theta = 0:180; % Projection angles

[R, xp] = radon(img, theta);

% Display the Radon transform

figure;

imagesc(theta, xp, R);

colormap(hot);

colorbar;

title('Radon Transform');

% Apply a filter (e.g., Ram-Lak) during back-projection

% Reconstruct the image using the inverse Radon transform (Filtered)

filtered_reconstruction = iradon(R, theta, 'linear', 'Ram-Lak', 1, size(img, 1));

% Reconstruct the image using the inverse Radon transform (Unfiltered)

unfiltered_reconstruction = iradon(R, theta, 'linear', 'none', 1, size(img, 1));

% Display the filtered reconstruction

figure;
 imshow(filtered_reconstruction, []);

title('Reconstructed Image (Filtered)');

% Display the unfiltered reconstruction

figure;

imshow(unfiltered_reconstruction, []);

title('Reconstructed Image (Unfiltered)');

% Compare results

figure;

subplot(1, 3, 1);

imshow(img, []);

title('Original Image');

subplot(1, 3, 2);

imshow(filtered_reconstruction, []);

title('Filtered Reconstruction');

subplot(1, 3, 3);

imshow(unfiltered_reconstruction, []);

title('Unfiltered Reconstruction');

Code: 2

% Read the brain MRI image

img = imread('Brain MRI.jpg');

% Convert to grayscale if it's RGB

if size(img, 3) == 3

img = rgb2gray(img);
end

% Resize the image to a manageable size for computation

img = imresize(img, [256, 256]);

% Define angles for analysis

angles = [50, 80, 120, 170];

% Initialize figure for comparison

figure;

for i = 1:length(angles)

% Perform the Radon transform for the specific angle

theta = 0:angles(i); % Projection angles up to the current angle

[R, xp] = radon(img, theta);

% Reconstruct the image using the inverse Radon transform (Filtered)

filtered_reconstruction = iradon(R, theta, 'linear', 'Ram-Lak', 1, size(img, 1));

% Reconstruct the image using the inverse Radon transform (Unfiltered)

unfiltered_reconstruction = iradon(R, theta, 'linear', 'none', 1, size(img, 1));

% Display results

subplot(length(angles), 3, (i-1)*3 + 1);

imagesc(theta, xp, R);

colormap(hot);

title(['Radon Transform (Angle ', num2str(angles(i)), ')']);

colorbar;

subplot(length(angles), 3, (i-1)*3 + 2);

imshow(filtered_reconstruction, []);
 title(['Filtered Reconstruction (Angle ', num2str(angles(i)), ')']);

subplot(length(angles), 3, (i-1)*3 + 3);

imshow(unfiltered_reconstruction, []);

title(['Unfiltered Reconstruction (Angle ', num2str(angles(i)), ')']);

end

Code Explanation:
Both code snippets aim to calculate the inverse Radon transform and compare filtered and unfiltered
back projections. The primary difference is how they approach the angle analysis and the way results are
presented.

Code 1: Single Angle Set (Full Range 0° to 180°)

1. Objective:
o Analyze the entire range of projection angles (0° to 180°) at once.
o Compare filtered and unfiltered back projections for a single comprehensive analysis.
2. Key Steps:
o Image Preprocessing: Load, grayscale conversion, resizing.
o Radon Transform: Performed for all angles from 0° to 180°.
o Reconstruction:
 Filtered using the Ram-Lak filter.
 Unfiltered reconstruction.
o Visualization:
 Compare the original image, filtered, and unfiltered reconstructions.
3. Best Use:
o Provides an overall reconstruction quality comparison across the full range of angles.
o Useful when you need a direct comparison of different reconstruction methods.

Code 2: Multi-Angle Analysis

1. Objective:
o Perform and compare Radon transforms and reconstructions at multiple subsets of
angles (e.g., 50°, 80°, etc.).
o Study how the number of angles affects the reconstruction quality.
2. Key Steps:
o Image Preprocessing: Load, grayscale conversion, resizing.
o Radon Transform: Iteratively performed for subsets of angles.
o Reconstruction:
 Filtered reconstruction using the Ram-Lak filter.
  Unfiltered reconstruction.
o Visualization:
 Display Radon transforms, filtered reconstructions, and unfiltered
reconstructions for each subset of angles.
3. Best Use:
o Ideal for analyzing the effect of angular coverage on reconstruction quality.
o Helps determine the minimum number of projection angles required for
acceptable reconstruction.

Comparison of Features:

Feature Code 1 Code 2

Angle Converge Full range (0° to 180°) Multiple subsets (e.g., 50°, 80°)

Focus General comparison of methods Effect of angle coverage

Reconstruction Filtered and unfiltered (global) Filtered and unfiltered (multi-case)

Visualization Simple (original, filtered, unfiltered) Detailed (per angle subset)

When to Use Each Code:

 Use Code 1 when:

o We need a straightforward comparison of reconstruction methods.
o Computational resources are limited (only one transform and reconstruction needed).
 Use Code 2 when:
o You're studying how angular resolution impacts reconstruction quality.
o You want detailed insights into the effects of angle subsets.

Potential Enhancements:

1. Code 1:
o Add metrics like mean squared error (MSE) or structural similarity index (SSIM)
for quantitative evaluation.
2. Code 2:
o Overlay original and reconstructed images to visualize discrepancies more clearly.
o Plot reconstruction error as a function of the number of angles.

Result:
Code: 1

Figure 1(Code 1 Script)

Figure 2(Original image, Reconstructed Image after filter, Reconstructed image without Filter)
 Figure 3(Result of Redon Transform at 180 degree)

Code: 2

Figure 4(Random Transform & inverse Random Transform at angle of 50, 80, 120 & 170 degree's)

Discussion:
While the results of the filtered back projection were much sharper and more accurate than the unfiltered
reconstruction, there are some considerations for real-world applications. The choice of filter plays a
significant role in the quality of the reconstructed image. Filters like Ram-Lak are commonly used, but
other filters may perform better depending on the specific application and the type of data being
reconstructed. Additionally, the lab focused on the basic principles of back projection, but more
advanced techniques, such as iterative reconstruction methods or artificial intelligence-based methods,
can further improve image quality, reduce artifacts, and handle noisy data more effectively.

Conclusion:
In this lab, I learned how to apply the Radon transform and the inverse Radon transform to reconstruct
images from projection data, specifically in the context of MRI image processing. The Radon transform
allows us to obtain projections of the original image at various angles, while the inverse Radon
transform (through back projection) helps us reconstruct the image from these projections. By applying
filtering techniques like the Ram-Lak filter during the back projection process, we can significantly
improve the quality of the reconstructed image, reducing blurring and enhancing sharpness.