Related Experiment Videos
Using unsupervised learning with variational bayesian mixture of factor analysis to identify patterns of glaucomatous
Pamela A Sample1, Kwokleung Chan, Catherine Boden
1Hamilton Glaucoma Center, Department of Ophthalmology, University of California-San Diego, La Jolla, 92093-0946, USA. psample@eyecenter.ucsd.edu
Investigative Ophthalmology & Visual Science
|July 28, 2004
Summary
An unsupervised machine learning classifier identified distinct visual field loss patterns in glaucoma patients. This method accurately grouped healthy eyes and categorized glaucomatous optic neuropathy (GON) visual fields, mirroring expert clinical experience.
Area of Science:
- Ophthalmology
- Machine Learning
- Medical Diagnostics
Background:
- Glaucomatous optic neuropathy (GON) is a leading cause of irreversible blindness.
- Accurate identification of visual field loss patterns is crucial for diagnosing and managing GON.
- Traditional methods rely on expert interpretation, which can be subjective and time-consuming.
Purpose of the Study:
- To evaluate an unsupervised machine learning classifier's ability to detect characteristic visual field loss patterns in GON.
- To determine if machine learning can replicate patterns identified through decades of human clinical experience.
Main Methods:
- A variational Bayesian mixture of factor analysis (vbMFA) unsupervised classifier was used.
- The classifier analyzed standard perimetry thresholds from 156 patients with GON and 189 healthy subjects.
- Data included 52 visual field locations and patient age.
Main Results:
- The vbMFA algorithm formed five distinct clusters based on visual field patterns.
- Cluster 5 predominantly contained normal eyes (98.4%), with some early GON cases.
- The remaining four clusters captured distinct patterns of GON-related visual field loss, including localized, superior, inferior, and bilateral defects.
- The classifier effectively separated healthy and GON eyes, with 70.5% of GON eyes falling into the four distinct defect clusters.
Conclusions:
- Unsupervised machine learning, specifically vbMFA, can identify clinically relevant patterns of visual field loss in GON.
- This approach aligns with expert clinical judgment and offers a data-driven method for pattern recognition in visual field testing.