Extracting medically relevant features from the human retina

Background: The Retina is “the area at the back of the eye that receives light and sends pictures of what the eye sees to the brain” (Cambridge Dictionary). It is the only part of our cardio vascular system that can be observed non-intrusively (without having the perform an operation): from looking at it, doctors can learn about our state of health regarding a range of diseases.

We would like to be able to use computers to automatically extract numerical values (features) from Retinal Images that could be of potential use for doctors.

Goal: The goal of this project is to implement a MATLAB program to manipulate retinal images with the objective of extracting numerical values that could be used to assess the state of health of an individual. A simple example of an interesting features could be the number of blood vessels in the retina. We will work on (anonymized) data from real patients from a Swiss Cohort related to hypertension (SKIPHOG).

Computational Tools: Students will be asked to write a MATLAB program to manipulate 2D retinal images. The task will be simplified by the use of a MATLAB library called “Image processing Toolbox”.

Mathematical tools: Students will be asked to use simple statistics (comparison of probability distributions) to assess whether the features they have extracted carry information. This will involve looking how features vary between the left and right eye of a single individual, vs. how the vary between different individuals. Indeed we argued that features that vary as much between the left and right eye of a subject, as the vary across subjects are unlikely to have a genetic (or environmental) component that asserts an effect to some but not other subjects and (assumedly) does so in both eyes to the same degree. The assumption is based on our belief that gene expression etc. can hardly be differential between the two eyes (unless something external, like an accident or a tutor broke the symmetry. While I think this is a reasonable argument, facial left-right symmetry is of course not perfect, and, interestingly it seems the degree of symmetry is an indicator of fitness (that's why facial symmetry, unless perfect, is often perceived as attractive). It could therefore be also quit interesting to find out if some of the individuals have more striking left-right (a)symmetry than others …

Biological or Medical aspects: the state of our retinal provides information about diseases such as hypertension, atherosclerosis, macular degeneration, retinopathy, glaucoma of the eye.

Supervisor: Mattia Tomasoni

Kick-off meeting presentation File:Extracting medically relevant features of the human retina 2018.pdf