Fuzzy clustering of 24-2 visual field patterns can detect glaucoma progression.

Area of Science:

• Ophthalmology
• Data Science
• Medical Imaging

Background:

• Visual field (VF) loss is a key indicator of glaucoma and other neurological conditions.
• Accurate representation and prediction of VF changes are crucial for patient management.
• Current methods may lack the granularity to capture individual VF patterns effectively.

Purpose of the Study:

• To develop and validate a hybrid archetypal analysis (AA) and fuzzy c-means (FCM) clustering approach for analyzing 24-2 visual field (VF) data.
• To represent individual patient VF losses using characteristic patterns.
• To enhance the detection and prediction of VF progression.

Main Methods:

• A multicenter retrospective study utilizing 132,938 VFs from 18,033 participants.
• Classification of 24-2 VF patterns using AA, followed by decomposition with FCM clustering.
• Supervised machine learning was employed to predict mean deviation (MD) change from decomposition coefficients.
• Comparison of Areas Under the Receiver Operating Characteristic Curves (AUCs) for detecting VF progression using different criteria.

Main Results:

• Identification of 16 characteristic VF patterns (archetypes) using AA.
• The hybrid AA-FCM approach demonstrated superior performance in predicting MD change compared to AA alone (lower mean squared error, higher correlation coefficient).
• FCM decomposition coefficient slopes showed improved detection of VF progression for specific defects (superior altitudinal, double arcuate, total loss) compared to AA.

Conclusions:

• A hybrid AA-FCM approach effectively visualizes 24-2 VF tests into characteristic patterns.
• This method enhances the detection of VF progression with lossless decomposition.
• The findings suggest a valuable tool for understanding and monitoring vision loss in clinical practice.