A Spatial Model of Visual Fields with Applications to Adaptive Sampling T Elze1, P Benner2, L Pasquale3, L Shen3, P Bex4 |
|---|
1Schepens Eye Research Institute, Harvard Medical School, MA, United States |
| Visual fields (VFs), the spatial array of perimetric sensitivity, are frequently assessed in vision research and ophthalmology to diagnose the functional loss related to diseases like glaucoma. In clinical practice, VFs are typically measured with automated perimeters that return sensitivities for a set of isolated locations, that ignore the spatial structure of VFs. In addition, the reliability of the VF test is only estimated by global indices, e.g. false positives/negatives, but not specified for individual locations. Here, we introduce a spatial probability model of VFs that transforms any set of discrete measurements into a continuous probability distribution that extends over the whole region of interest. The model includes a measure of credibility for each VF location and takes into account the noise distribution at each location and the connectivity strength among locations in the VF. These parameters can be used online to increase the efficiency of adaptive testing. Our model is designed to be used for medical diagnoses over the ratio of marginal likelihoods (Bayes Factors) and for the quantification of VF loss progression over the Kullback-Leibler divergence. The model can specifically address the diagnosis of different eye diseases. We show an application to glaucoma as a proof of concept. |
| Up Home |
|---|