Glaucoma, one of the leading causes of blindness, often develops asymptomatically, necessitating early diagnosis and prediction of the progression rate of glaucomatous optic neuropathy (GON).

PURPOSE

To develop a classification model using machine learning methods for predicting the rate of GON progression, and to identify the most significant predictors of progression in patients with newly diagnosed early primary open-angle glaucoma (POAG).

MATERIAL AND METHODS

The study included 59 patients (59 eyes) with early POAG, categorized into three groups based on the expert assessment of GON progression rate over a 36-month follow-up using dynamic morphofunctional evaluation. A classification model incorporating 35 clinical parameters, including optical coherence tomography (OCT) and OCT-angiography (OCT-A) data, was developed using partial least squares discriminant analysis (PLS-DA).

RESULTS

Over the 36-month follow-up, slow GON progression was recorded in 21 patients, moderate in 18, and rapid in 20. The mean progression rates were −0.77±1.27%/year for visual field area, −1.21±1.48 µm/year for retinal nerve fiber layer (RNFL) thickness, and −1.23±1.77 µm/year for ganglion cell complex (GCC) thickness. The model demonstrated sensitivity of 90%, specificity of 95%, and efficiency of 92%. The most significant predictors of GON progression were mean vessel density in the deep vascular plexus of the macular region (wiVD_Deep), choriocapillaris dropout in the inferior-nasal peripapillary region, choroidal thickness in the fovea, and lamina cribrosa thickness.

CONCLUSION

The developed model effectively classifies patients based on the predicted progression rate of GON, which is important for individualized approach to glaucoma treatment planning.