Abstract
Purpose :
Without proper protection, the eye is especially vulnerable to sharp objects
that can penetrate and damage the retina in the posterior segment. A posterior
penetrating eye injury can lead to retinal tearing, intraocular fibrosis and retinal
detachment that require timely intervention without compromising vision. Currently,
ophthalmoscopy is the gold standard for visualizing the retina, even though one cannot
consistently detect retinal abnormalities by relying on such a method alone. The
objective of this work was to determine whether automated image processing systems
based on machine learning could facilitate the identification of subtle changes in the
retina following an injury.
Methods :
Ten male Dutch Belted rabbits were subjected to posterior penetrating eye
injury in the right eye using a 23-gauge needle. Injury progression was monitored via
fundus photography at weekly intervals over a period of 4 weeks. The dataset consisted
of 743 full-color fundus images that were labelled as injured or uninjured and randomly
split into training and testing groups at a ratio of 7:3. Two types of machine learning
models, convolution neural network (CNN) and support vector machine (SVM), were
constructed using training group images and validated with testing group images. The
open-source Keras Python library was adapted to generate a CNN model, whereas the
scikit-learn Python library was implemented to create four SVM models. The models
were compared using measures such as accuracy, precision, and the area under the
receiver operating characteristic curve (ROC AUC).
Results :
Most models exhibited promising accuracy for the classification of full-color
fundus images. The CNN model had relatively high accuracy of 91.03% and ROC AUC
of 0.91. The SVM polynomial kernel performed the best across all metrics, with 92.37%
accuracy and ROC AUC of 0.96. The SVM linear kernel presented comparable but
slightly lower metrics. The worst performing models were the SVM sigmoid and radial
basis function kernels.
Conclusions :
Our study shows that trained machine learning algorithms can accurately
identify changes in the retina following trauma to the posterior segment of the eye. This
is an important step towards developing computer-aided diagnostic tools that can be
used to detect injury and disease progression following posterior penetrating eye injury.
This abstract was presented at the 2024 ARVO Imaging in the Eye Conference, held in Seattle, WA, May 4, 2024.