DIAGNOSE EYES DISEASES USING VARIOUS FEATURES EXTRACTION APPROACHES AND MACHINE LEARNING ALGORITHMS

Authors

  • Zahraa Najm Abed Iraqi Commission for Computers and Informatics (ICCI)
  • Abbas M Al-Bakry University of Information Technology and Communications

DOI:

https://doi.org/10.25195/ijci.v49i2.437

Keywords:

Machine Learning, Random Forest, Naïve Bayes, K-nearest Neighbor, ODIR

Abstract

Ophthalmic diseases like glaucoma, diabetic retinopathy, and cataracts are the main cause of visual impairment worldwide. With the use of the fundus images, it could be difficult for a clinician to detect eye diseases early enough. By other hand, the diagnoses of eye disease are prone to errors, challenging and labor-intensive. Thus, for the purpose of identifying various eye problems with the use of the fundus images, a system of automated ocular disease detection with computer-assisted tools is needed. Due to machine learning (ML) algorithms' advanced skills for image classification, this kind of system is feasible. An essential area of artificial intelligence)AI (is machine learning. Ophthalmologists will soon be able to deliver accurate diagnoses and support individualized healthcare thanks to the general capacity of machine learning to automatically identify, find, and grade pathological aspects in ocular disorders. This work presents a ML-based method for targeted ocular detection. The Ocular Disease Intelligent Recognition (ODIR) dataset, which includes 5,000 images of 8 different fundus types, was classified using machine learning methods. Various ocular diseases are represented by these classes. In this study, the dataset was divided into 70% training data and 30% test data, and preprocessing operations were performed on all images starting from color image conversion to grayscale, histogram equalization, BLUR, and resizing operation. The feature extraction represents the next phase in this study ,two algorithms are applied to perform the extraction of features which includes: SIFT(Scale-invariant feature transform) and GLCM(Gray Level Co-occurrence Matrix), ODIR dataset is then subjected to the classification techniques Naïve Bayes, Decision Tree, Random Forest, and K-nearest Neighbor. This study achieved the highest accuracy for binary classification (abnormal and normal) which is 75% (NB algorithm), 62% (RF algorithm), 53% (KNN algorithm), 51% (DT algorithm) and achieved the highest accuracy for multiclass classification (types of eye diseases) which is 88% (RF algorithm), 61% (KNN algorithm) 42% (NB algorithm), and 39% (DT algorithm).

Downloads

Download data is not yet available.

Author Biography

Zahraa Najm Abed, Iraqi Commission for Computers and Informatics (ICCI)

Informatics Institute for Postgraduate Studies

Downloads

Published

2023-09-30