Related Experiment Videos
Artificial neural network analysis of noisy visual field data in glaucoma
D B Henson1, S E Spenceley, D R Bull
1Department of Ophthalmology, University of Manchester, Royal Eye Hospital, UK. david.henson@man.ac.uk
Artificial Intelligence in Medicine
|June 1, 1997
Summary
This study uses a self-organizing map (SOM) artificial neural network to analyze glaucoma visual field data, effectively handling noise for better longitudinal assessment and diagnosis.
Area of Science:
- Ophthalmology
- Artificial Intelligence
- Data Science
Background:
- Glaucoma diagnosis relies on longitudinal visual field analysis, which is often complicated by data noise.
- Detecting significant changes in visual fields is crucial for clinical diagnosis but challenging due to inherent variability.
Purpose of the Study:
- To apply a 2-dimensional Kohonen self-organizing map (SOM) for analyzing ophthalmological data, specifically visual fields of glaucoma patients.
- To address the challenge of noise within longitudinal visual field data for improved clinical analysis.
- To enhance the visual representation and quantification of significant class change in longitudinal glaucoma data.
Main Methods:
- Utilized a 2-dimensional Kohonen self-organizing map (SOM) with 25 nodes on a square grid to analyze 737 glaucomatous visual field records.
- Trained the SOM to cluster visual field data, positioning early and advanced loss at extreme nodes with a continuum of change in between.
- Incorporated noise analysis by classifying 100 simulated variants for each SOM node to establish classification noise extent.
Main Results:
- The SOM successfully clustered glaucomatous visual field data, creating a spatial map representing disease progression.
- Noise analysis quantified the impact of data variability on classification, allowing for more robust interpretation.
- Field change was measured by comparing subsequent fields against original fields and their simulated variants, enabling significant change detection.
Conclusions:
- The application of SOM for spatial analysis of visual field data, augmented with noise analysis, enhances the visual representation of longitudinal glaucoma data.
- This approach enables more accurate quantification of significant class change, improving clinical diagnosis and patient monitoring.
- The study demonstrates the utility of SOM in managing noisy ophthalmological data for more reliable longitudinal assessment.