Patterns of glaucomatous visual field loss in sita fields automatically identified using independent component analysis.

PURPOSE To determine if the patterns uncovered with variational Bayesian-independent component analysis-mixture model (VIM) applied to a large set of normal and glaucomatous fields obtained with the Swedish Interactive Thresholding Algorithm (SITA) are distinct, recognizable, and useful for modeling the severity of the field loss. METHODS SITA fields were obtained with the Humphrey Visual Field Analyzer (Carl Zeiss Meditec, Inc, Dublin, California) on 1,146 normal eyes and 939 glaucoma eyes from subjects followed by the Diagnostic Innovations in Glaucoma Study and the African Descent and Glaucoma Evaluation Study. VIM modifies independent component analysis (ICA) to develop separate sets of ICA axes in the cluster of normal fields and the 2 clusters of abnormal fields. Of 360 models, the model with the best separation of normal and glaucomatous fields was chosen for creating the maximally independent axes. Grayscale displays of fields generated by VIM on each axis were compared. SITA fields most closely associated with each axis and displayed in grayscale were evaluated for consistency of pattern at all severities. RESULTS The best VIM model had 3 clusters. Cluster 1 (1,193) was mostly normal (1,089, 95% specificity) and had 2 axes. Cluster 2 (596) contained mildly abnormal fields (513) and 2 axes; cluster 3 (323) held mostly moderately to severely abnormal fields (322) and 5 axes. Sensitivity for clusters 2 and 3 combined was 88.9%. The VIM-generated field patterns differed from each other and resembled glaucomatous defects (eg, nasal step, arcuate, temporal wedge). SITA fields assigned to an axis resembled each other and the VIM-generated patterns for that axis. Pattern severity increased in the positive direction of each axis by expansion or deepening of the axis pattern. CONCLUSIONS VIM worked well on SITA fields, separating them into distinctly different yet recognizable patterns of glaucomatous field defects. The axis and pattern properties make VIM a good candidate as a preliminary process for detecting progression.

[1]  M H Goldbaum,et al.  Interpretation of automated perimetry for glaucoma by neural network. , 1994, Investigative ophthalmology & visual science.

[2]  H. Rootzén,et al.  A new generation of algorithms for computerized threshold perimetry, SITA. , 2009, Acta ophthalmologica Scandinavica.

[3]  Sejnowski,et al.  ICA MIXTURE MODELS FOR UNSUPERVISED CLASSIFICATION AND AUTOMATIC CONTEXT SWITCHING , 2000 .

[4]  Terrence J. Sejnowski,et al.  Variational Learning of Clusters of Undercomplete Nonsymmetric Independent Components , 2003, J. Mach. Learn. Res..

[5]  Robert N Weinreb,et al.  Comparing machine learning classifiers for diagnosing glaucoma from standard automated perimetry. , 2002, Investigative ophthalmology & visual science.

[6]  Robert N Weinreb,et al.  Using machine learning classifiers to identify glaucomatous change earlier in standard visual fields. , 2002, Investigative ophthalmology & visual science.

[7]  Robert N Weinreb,et al.  Using unsupervised learning with variational bayesian mixture of factor analysis to identify patterns of glaucomatous visual field defects. , 2004, Investigative ophthalmology & visual science.

[8]  Robert N Weinreb,et al.  Using unsupervised learning with independent component analysis to identify patterns of glaucomatous visual field defects. , 2005, Investigative ophthalmology & visual science.

[9]  Michael Henry Goldbaum,et al.  Unsupervised learning with independent component analysis can identify patterns of glaucomatous visual field defects. , 2006, Transactions of the American Ophthalmological Society.

[10]  T. Sejnowski,et al.  Unsupervised machine learning with independent component analysis to identify areas of progression in glaucomatous visual fields. , 2005, Investigative ophthalmology & visual science.

[11]  Christopher Bowd,et al.  Unsupervised Machine Learning With Independent Component Analysis Identifies Patterns of Glaucomatous Visual Field Loss in SITA Fields , 2009 .