Visual field prediction using K-means clustering in patients with primary open angle glaucoma.

Area of Science:

• Ophthalmology
• Medical Imaging
• Data Science

Background:

• Primary open angle glaucoma (POAG) is a leading cause of irreversible blindness.
• Accurate long-term prediction of visual field (VF) progression is crucial for managing POAG.
• Current prediction methods may have limitations, especially with sparse data.

Purpose of the Study:

• To evaluate the efficacy of K-means clustering for predicting long-term visual field outcomes in POAG patients.
• To compare the predictive accuracy of K-means clustering against other methods like hierarchical ordered partitioning and collapsing hybrid (HOPACH) and pointwise linear regression (PLR).

Main Methods:

• Utilized data from 228 POAG patients (training) and 81 patients (testing) with at least 10 visual field tests (24-2 VF).
• Applied K-means clustering and HOPACH to cluster 52 total deviation values (TDVs) from initial VF tests.
• Employed linear regression on clustered regions to predict TDVs of the 10th VF test, comparing prediction errors (RMSE).

Main Results:

• K-means clustering identified nine distinct VF regions, while HOPACH identified eleven.
• K-means clustering demonstrated significantly lower prediction errors than PLR for prediction ratios of 1:3 and 1:4.
• K-means clustering generally outperformed HOPACH in prediction accuracy across various ratios, except for 1:3.

Conclusions:

• K-means clustering provides a more accurate method for predicting long-term visual field test results in POAG patients.
• This clustering approach is particularly beneficial when dealing with limited visual field data.
• K-means clustering shows promise as a valuable tool for enhancing glaucoma management strategies.