Retinal Blood Vessel Segmentation

Problem

The code aims to perform retinal blood vessel segmentation using various image processing techniques. Additionally, it includes a validation function to assess the accuracy of the segmentation results compared to ground truth data.

Mathematical model

• Image Processing Techniques: The code implements several image processing steps to segment blood vessels by using Filtering and Thresholding Techniques like Gaussian filtering, erosion (using a diamond-shaped structuring element), principal curvature calculation, contrast enhancement (histogram equalization), and thresholding are used to identify blood vessels.
• Validation Technique: The validation function compares the segmented image with a ground truth image to compute accuracy metrics in the training phase.

Selected methods

1. Structuring Element and Erosion: A diamond-shaped structuring element (strel('diamond', 20)) is applied for erosion to refine the mask.
2. Gaussian Filter: Gaussian filtering (imgaussfilt) with a sigma value of 1.45 is used to smoothen the green channel of the input image.
3. Principal Curvature Calculation: A function prinCur calculates the principal curvature of the filtered image.
4. Histogram Equalization and Thresholding: Adaptive histogram equalization (adapthisteq) is performed on the calculated curvature. Thresholding (isodata) is used to segment the vessels.

Test cases Demonstrations

1- Training Phase:

• Reads training images.
• Processes each image using the described pipeline.
• Generates segmented images and saves them in the processed folder.
• Displays intermediate processing steps if the mode is 'train'.

• The validation function assesses segmentation accuracy by comparing the segmented image with ground truth data. It calculates metrics like True Positive Rate (TPR), False Positive Rate (FPR), and Accuracy (AC).

2- Testing Phase:

• Reads test images.
• Applies the same processing pipeline.
• Saves the resulting segmented images in the results folder.
• Displays intermediate steps if the mode is 'test' (commented out in the code).
• There is no validation here as I don’t have ground truth values in the test folder database to compare with.