To remove background noise from MRI images in MATLAB

by Posted on

To remove background noise from MRI images in MATLAB, follow these steps using thresholding, morphological operations, and connected component analysis:

Step-by-Step Solution:

  1. Read the MRI Image:
    Load the image, convert to grayscale if necessary, and prepare it for processing.
  2. Thresholding:
    Use Otsu’s method to automatically determine the optimal threshold to separate foreground (anatomy) from background noise.
  3. Morphological Cleaning:
    Remove small noise artifacts and fill holes in the mask using morphological operations.
  4. Largest Connected Component:
    Identify the main anatomical structure by selecting the largest connected region in the mask.
  5. Apply Mask:
    Multiply the original image by the cleaned mask to set background pixels to zero.

MATLAB Code:

matlab
Copy
% Read the MRI image (replace with your image)
mri_image = imread('mri_image.jpg'); % Use dicomread for DICOM files

% Convert to grayscale if RGB
if size(mri_image, 3) == 3
    mri_image = rgb2gray(mri_image);
end

% Convert to double for processing and normalize
image_double = im2double(mri_image);

% Step 1: Apply Gaussian smoothing to improve thresholding
smoothed_image = imgaussfilt(image_double, 2);

% Step 2: Calculate Otsu's threshold and create initial mask
level = graythresh(smoothed_image);
mask = imbinarize(smoothed_image, level);

% Step 3: Clean the mask using morphological operations
mask = imopen(mask, strel('disk', 5)); % Remove small noise
mask = imfill(mask, 'holes'); % Fill holes

% Step 4: Keep the largest connected component
cc = bwconncomp(mask);
[~, idx] = max(cellfun(@numel, cc.PixelIdxList));
mask = false(size(mask));
mask(cc.PixelIdxList{idx}) = true;

% Step 5: Apply the mask to the original image
denoised_image = mri_image .* cast(mask, class(mri_image));

% Display results
figure;
subplot(1,2,1), imshow(mri_image, []), title('Original Image');
subplot(1,2,2), imshow(denoised_image, []), title('Background Removed');

Explanation:

  • Thresholding: Otsu’s method (graythresh) separates foreground and background by maximizing inter-class variance.
  • Morphological Operations: imopen removes small noise, and imfill fills gaps in the mask.
  • Largest Component: Ensures only the main anatomical structure is retained.
  • Mask Application: Sets background pixels to zero, effectively removing noise.

Notes:

  • DICOM Images: Replace imread with dicomread for DICOM files.
  • Parameter Tuning: Adjust the disk size in imopen based on noise level.
  • Advanced Cases: For non-uniform backgrounds or intensity bias, consider advanced methods like active contours or bias correction.

This approach effectively isolates the anatomical structure and eliminates background noise.

Published by howdy