Glaucoma progression detection using structural retinal nerve fiber layer measurements and functional visual field points

Abstract

Machine learning classifiers were employed to detect glaucomatous progression using longitudinal series of structural data extracted from retinal nerve fiber layer thickness measurements and visual functional data recorded from standard automated perimetry tests. Using the collected data, a longitudinal feature vector was created for each patient's eye by computing the norm 1 difference vector of the data at the baseline and at each follow-up visit. The longitudinal features from each patient's eye were then fed to the machine learning classifier to classify each eye as stable or progressed over time. This study was performed using several machine learning classifiers including Bayesian, Lazy, Meta, and Tree, composing different families. Combinations of structural and functional features were selected and ranked to determine the relative effectiveness of each feature. Finally, the outcomes of the classifiers were assessed by several performance metrics and the effectiveness of structural and functional features were analyzed.

Fig. 1

Top left: Sample optic disc photograph image, Top right: SAP visual points tested using the 24-2 system, and bottom: Sample OCT RNFL image and its scan type (circular optical section around the optic nerve).

Fig. 2

(a) Data vector formation, (b) longitudinal data formation for each subject’s eye, (c) feature vector generation.

Fig. 3

ROC curves for different feature sets; (a) Combined features from all RNFL and all SAP, (b) all RNFL features only, (c) all SAP features only, and (d) the best classifier (based on AUROC) for combined features from all RNFL and all SAP, all RNFL features only, and all SAP features only.

Fig. 4

ROC curves for different classifiers when we use only top ten ranked features.

Fig. 5

Probability assignment to longitudinal visit of each patient. (a) Stable group. (b) Progressing group. Connected circles in the same color indicate the classifier outcome for the same subject’s eye longitudinal data.

Fig. 6

Feature distributions and Bayesian discriminating function plotted in 2-D for two best features. Each blue circle represents the RNFL inferior temporal versus inferior nasal feature (difference between RNFL thickness at baseline and follow-up in micron) at a follow-up of an eye that belongs to the stable group, and each orange circle represents RNFL inferior temporal versus inferior nasal feature of an eye that belongs to the progressing group.

Fig. 7

Percent probability of progression based on Bayesian Net classifier outcome.

Fig. 8

Evolution of features over time (visits); (a) RNFL feature at inferior nasal sector for ten stable eyes over the time course of follow ups, (b) RNFL feature at inferior nasal sector for ten progressing eyes over the time course of follow ups. (Connected circles belong to a single eye and each circle represents the feature value at that visit, sorted from first follow-up to the last follow-up from left to right).