Early detection and diagnosis of colour blindness using machine learning
Loading...
Supplementary material
Other Title
Authors
Don, Reshla Dinali Wisidagamage
Author ORCID Profiles (clickable)
Degree
Master of Applied Technologies (Computing)
Grantor
Unitec, Te Pūkenga – New Zealand Institute of Skills and Technology
Date
2025
Supervisors
Sharifzadeh, Hamid
Keivanmarz, Ali
Keivanmarz, Ali
Type
Masters Thesis
Ngā Upoko Tukutuku (Māori subject headings)
Keyword
colour blindness
diagnosis
machine learning
pattern recognition systems in medicine
diagnosis
machine learning
pattern recognition systems in medicine
ANZSRC Field of Research Code (2020)
Citation
Don, R.D.W. (2024). Early detection and diagnosis of colour blindness using machine learning (Unpublished document submitted in partial fulfilment of the requirements for the degree of Master of Applied Technologies (Computing)). Unitec, Te Pūkenga - New Zealand Institute of Skills and Technology
https://hdl.handle.net/10652/6796
Abstract
RESEARCH QUESTIONS
• How the problem of insufficient fundus images will be addressed, and the methods used to normalise the fundus images in the two datasets?
• What is the most appropriate Machine Learning algorithm in diagnosing CVD eyes of an individual in terms of the model accuracy?
• How the severity of colour blindness will be measured and the anticipated thresholds that could be applied to analyse the severity of the CVD?
ABSTRACT
The technology and machine learning have grown to be an essential component in medical and healthcare domain, that transform medical procedures, operational efficiency, and patient care. In human body, eye plays an important role in vision and communication. Healthy eyes will lead to a clear perception and will act as the most substantial sense in human interaction and decision making. The cone cells in eye retina influence the sensitivity to different colours in different wavelengths. Any variation in this cone cells will result in a condition called Colour Vision Deficiency (CVD). This condition is further classified into three categories: Protanopia, Deuteranopia, Tritanopia and will be discussed throughout the thesis.
Diagnosing colour blindness has been conducted in various medical and computational investigations. However, these methods have proven to be time consuming, require high computational resources, and additional support. Additionally, it lacks machine learning integration, specifically in diagnosing CVD. In this thesis, we are discussing about machine learning and deep learning mechanisms to pre diagnose CVD as a self-assessment tool. This is performed as a multi classification supervised learning task to accurately classify the eye fundus images of normal vision, colourblind vision, and other types of diseases. The multi classification is modelled using Convolutional Neural Networks (CNN), Support Vector Machine (SVM), and Random Forest (RF) to test and identify the best fitting model to classify individual eye fundus efficiently. All three algorithms are trained and tested using a combined dataset of eye fundus images. Based on the results and evaluation of the three models and its accuracy metric, the CNN model is proposed as the optimal solution for identifying CVD in individuals with an accuracy of 99.71%. Following the classification of colourblind eye fundus, the severity is measured using Mean Squared Error (MSE) by comparing a healthy eye fundus with colourblind versions of eye fundus. This is calculated as a quantitative measurement to determine the severity of CVD of an individual. Consequently, the quantitative result is transformed to a qualitative measurement by setting up thresholds of mild, medium, and severe levels. The severity thresholds are set by following the guidelines of the Colour Assessment and Diagnostic (CAD) and its Severity Index (SI) which explains that the levels could be set up depending on the requirements of the specific investigation.
Publisher
Permanent link
Link to ePress publication
DOI
Copyright holder
Author
Copyright notice
All rights reserved