Retinal components detection method

An algorithm destined to identify, separate and segment the main components of the retina using fundus images. This algorithm is created using Python and OpenCv for the image processing.

This method is composed by three sub-methods, which focus on detecting and segmenting each of the main anatomical components found in the retina. The three sub-methods found in the algorithm are as follows:

1. Vascular tree detection and segmentation

The detection of the vascular tree in the retina using a Random Forest model in order to do the classification between the non-vascular elements and the vascular elements. The model was trained using the DRIVE datasets for the supervised training reference. This method is based on the work directed by (GeethaRamani and Balasubramanian,2016). ¹

2. Optic Disc detection and segmentation

The detection of the Optic Disc structure in the retina using a voting mechanism based on statistical analysis and color intensity analysis. This method is based on the work directed by (Aquino et al., 2010). ²

3. Macula detection and segmentation

The detection of the Macula component based on color and morphological analysis on high contrast elements on the retina. This method is based on the work directed by (GeethaRamani and Balasubramanian,2018). ³

Each of the sub-methods is developed in its own .py file, found in the master branch, and are combined in the class file ret_fundus.py in order obtain the retinal elements and segment each of them. The result is an object with a dictionary structure of each element containing its segmented mask, center location and radius for the circular elements.

Datasets used for this project

The datasets used for this project is a recopilation of sets of accepted images from different datasets. The complete datasets are found here, while the public datasets are found as follows:

• DRIVE datasets

Footnotes and further considerations

To-do list:

• Set a functional detection and segmentation method for the anatomical components in the retina.
• Find a way to save the segmented components in a data structure for future references.
• Correct the algorithm to detect pathological elements in the fundus images.
• Complete the anatomical estimation search for the vascular arch and the fovea location estimation.
• Set up the database architecture for resulting structures storing and display.
• Generate a GUI application for display and control of the resulting algorithms.

Footnotes:

Footnotes

1. Retinal blood vessel segmentation employing image processing and data mining techniques for computerized retinal image analysis ↩

2. Detecting the optic disc boundary in digital fundus images usingmorphological, edge detection, andfeature extraction techniques ↩

3. Macula segmentation and fovealocalization employing image processing and heuristic basedclustering for automated retinalscreening ↩