Related Experiment Videos
Using unsupervised learning with independent component analysis to identify patterns of glaucomatous visual field
Michael H Goldbaum1, Pamela A Sample, Zuohua Zhang
1Ophthalmic Informatics Laboratory and Hamilton Glaucoma Center, Department of Ophthalmology, University of California at San Diego, La Jolla, 92093, USA. mgoldbaum@ucsd.edu
Investigative Ophthalmology & Visual Science
|September 28, 2005
Summary
Unsupervised learning using variational Bayesian independent component analysis mixture model (vB-ICA-mm) effectively segmented glaucoma patterns in standard automated perimetry (SAP) data. This method identified distinct clusters and severity axes, proving valuable for glaucoma analysis.
Area of Science:
- Ophthalmology
- Machine Learning
- Data Mining
Background:
- Previous studies utilized machine learning classifiers for glaucoma pattern segmentation in standard automated perimetry (SAP).
- Unsupervised learning offers a method to decompose complex visual field data into meaningful components.
Purpose of the Study:
- To evaluate the information represented by axes derived from independent component analysis (ICA) of SAP field patterns.
- To apply variational Bayesian independent component analysis mixture model (vB-ICA-mm) for segmenting glaucoma patterns in SAP data.
Main Methods:
- SAP data from 189 normal eyes and 156 eyes with glaucomatous optic neuropathy (GON) were analyzed.
- The vB-ICA-mm model partitioned SAP fields into clusters and learned maximally independent axes for each cluster.
- Eyes were classified into 'Glaucoma' (G) or 'Normal' (N) clusters based on vB-ICA-mm analysis.
Main Results:
- Two informative clusters were identified: Cluster G (68.6% GON eyes) and Cluster N (98.4% normal eyes).
- Cluster G optimally contained six axes, with patterns at -1 SD and +2 SD correlating with expert-identified glaucoma defects.
- Increasing severity of SAP field defects correlated with the position along the positive direction of the cluster G mean axes.
Conclusions:
- vB-ICA-mm successfully represented SAP fields with patterns interpretable by glaucoma experts.
- The model captured the severity of visual field defects within the identified patterns.
- vB-ICA-mm is validated as a data-mining technique for complex, novel tests in ophthalmology.