Welcome to my portfolio site. Feel free the take a look around.

Automatic Detection of Retinal Vascular Bifurcations

Desktop Applications April 2015

Automatic detection for retinal vascular bifurcations is the subject of my 4th year Masters of Computing project. Below is a snippet taken from my final year dissertation with the aim to give a brief insight into the project, rationale and more information about the topic.

The main aim of this project is to automate the detection of vascular bifurcations within retinal fundus imagery, using methods researched throughout the course of this project. The process of detecting bifurcation points will be automated by developing a software algorithm, which takes the retinal fundus image as an input, performs some analysis techniques to locate the interested points within the image, then shows the results back to the user in the form of plotting the points onto the original input image. A bifurcation is a junction whereby one blood vessel splits to form 2 separate blood channels. This can be seen in the figure below. There are many different computer vision problems when trying to automatically detection these points, making it a current research topic. At the time of writing this work, there has been no full proof method of localising bifurcations found. The method proposed within this work has not been tried before.

Screen Shot 2015-05-17 at 10.48.25

2 retinal vascular bifurcations

This paper presents a new method of retinal vascular bifurcation localisation using intensity values and gradient changes. This method can be broken down into 6 different stages. Firstly the green layer from the RGB input image (an image from the DRIVE dataset) is extracted and normalised. Next the system builds a set of nested VShape templates with different angles using a specified length and shift between the inner and outer template. Then each of the two templates are applied individually and a ratio calculated to give an indication of the gradient change. The image is cleaned using thresholding and some morphological operations. This produces a binary image which can be used to locate the bifurcation points (taking into account a bifurcation point can be within a 5 pixel radius of the ground truth value) and plot them onto the original image. The last stage involves using machine learning to find the most appropriate template length and shift values as well as the threshold value used for generating a binary image. The results from this project can be found within section 8 of this report.

Automatic detection of retinal vascular bifurcations has many possible applications and benefits. For example, changes within the vascular network can indicate hypertension, diabetic retinopathy or coronary heart disease. One of the early stages in performing such a diagnosis is the detection of the bifurcation and cross over points. Currently this process is being performed manually, however, with over 100 vascular bifurcations in each typical retinal fundus image, the process is very time consuming. As a result, an automated method would not only be ideal but also allow such examinations to be completed more regularly.

In addition to the medical advantages, this work also has additional applications. One example would be retinal scans for authentication. Locating the specific bifurcation and crossover points for each individual would allow a security system to generate a map of that person’s retina network to be later used as an authentication mechanism.

Below is a gallery contains some of the outputs from this system. All marks on the image have been produced by the system. Red marks are false positive results, meaning the system thinks there is a bifurcation however the ground truth/gold standard tells us there is not. Green marks are true positive results, meaning the system thinks there is a bifurcation there and the ground truth/gold standard confirms that to be true.

This content has been taken from the assessment documentation created as part of this module.

This was part of my Computer Vision and Robotics module for year 4 for my Masters of Computing (MCOMP) degree with The University of Lincoln. If you would like a copy any documentation (including the corresponding report), then please contact me using the facilities on this website.



Browse portfolio


  • The University of Lincoln