Progression of patterns (POP): a machine classifier algorithm to identify glaucoma progression in visual fields

Area of Science:

• Ophthalmology
• Medical Technology
• Machine Learning in Healthcare

Background:

• Glaucomatous visual field (VF) defects can be challenging to detect early.
• Accurate identification of VF progression is crucial for timely intervention.

Purpose of the Study:

• To evaluate the efficacy of Progression of Patterns (POP) in identifying glaucomatous visual field (VF) defects.
• To compare POP's performance against established methods like Visual Field Index (VFI), Mean Deviation (MD), and Guided Progression Analysis (GPA).

Main Methods:

• POP utilizes a variational Bayesian independent component mixture model (VIM) machine learning classifier.
• VIM analyzed Swedish Interactive Thresholding Algorithm (SITA) VFs, identifying seven glaucomatous patterns.
• Progression was assessed using linear regression slopes on VIM axes, establishing a degree of confidence (DOC) for progression detection.

Main Results:

• POP identified a similar number of progressing eyes as VFI, MD, and GPA in suspect cases.
• POP identified significantly more progressing eyes than GPA in both glaucoma and progressive glaucomatous optic neuropathy (PGON) cohorts.
• Statistical analysis confirmed POP's superior performance (P=0.01 for glaucoma, P=0.05 for PGON) compared to GPA.

Conclusions:

• POP provides valuable information on the degree of confidence (DOC) for VF progression.
• POP's ability to identify specific progressing VF defect patterns enhances clinical decision-making.
• POP offers a valuable tool for clinicians in detecting VF progression in glaucoma patients.