Simultaneous modelling and clustering of visual field data

Abstract

Visual Field (VF) tests and their corresponding data are commonly used in clinical practices to manage glaucoma. The data represents patient visual acuity, which determines whether the patient has good or impaired vision. Developing machine learning and data mining algorithms that explore the spatial and temporal aspects of visual filed data could vastly improve early diagnosis as well as assisting practitioners in providing appropriate treatments. The objective of this study is to explore the simultaneous modelling and clustering of VF data so that a better understanding of the relationship between VF points can be made, as well as the generation of models that can better predict glaucoma progression. The spatial clusters over the visual field are determined by using heuristic search techniques which are scored based upon the prediction accuracy of glaucoma deterioration. This is compared to methods using standard clusters that are based upon physiological traits (the six optic nerve fiber bundles). Our results demonstrate an improvement in prediction accuracy for some of the models.