The present research compared the accuracy of machine learning algorithms in classifying the thickness and volume measurements of retinal layers as obtained from male and female subjects. The study evaluated the retina of sixty-four healthy participants (38 women and 26 men), with normal vision and without eye or systemic diseases, aged between 20 and 40 years. The data acquisition was obtained with a Spectralis HRA+OCT tomograph in the macular region of the retina and its layers: retinal nerve fiber layer (RNFL), ganglion cell layer (GCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), retinal pigment epithelium (RPE), inner retina (IRL) and outer retina (ORL). The classification accuracy was obtained with the following algorithms: support vector classifier (SVC), logistic regression (LR), linear discriminant analyses (LDA), k-nearest neighbors (kNN), decision tree (DT), gaussian naive bayes (GNB) and random forest (RF). The characteristics attributed to each participant's samples were the thickness values in the nine regions of the macula plus the total macular volume of each retinal layer. The statistical tests Two-way ANOVA and Tukey HSD post-hoc were used in the statistical comparisons between the accuracies for the classifier and retinal layer variables, considering a significance level of < 0.05. All factors (classifier, retinal layer, and their interactions) had significant influences on accuracy (p < 0.05). The main effect of the algorithm type factor resulted in an F ratio of F (6, 630) = 4.527, p = 0.0002. The main effect for the retinal layer produced an F ratio of F (9, 630) = 51.64 and p < 0.0001. The interaction effect was also significant, F(54, 630) = 1.741, p = 0.0012. All algorithms classified with high accuracy (> 0.70) the innermost layers of the retina (total retina, inner retina, RNFL, GCL, INL) according to the gender of the participants, where we observed significant differences between genders in thickness and measurements volume. The SVC, LDA, and LR algorithms produced high accuracy (>0.70) when thickness and volume data came from the RNFL compared to the outermost layers of the retina. The KNN, RF and DT algorithms performed better in correctly classifying the total retina data in relation to the outermost layers. The thickness and volume of the retina and the innermost layers of the retina allow machine learning algorithms to be more accurate in separating data from different sexes.
|
