Summary Current assessments of visual field testing depend on algorithms, principally developed to diagnose and monitor progression in glaucoma, or on expert descriptive categorization of deficits. The algorithms do not work well for non-glaucomatous optic neuropathies as these disorders can both improve and deteriorate. Descriptive categorizations are not readily quantifiable to assess change over time. Unsupervised statistical learning archetypal analysis is a new way to investigate glaucoma and potentially other optic neuropathies. Both idiopathic intracranial hypertension and optic neuritis are disorders that often improve and respond to therapy. Archetypal analysis of the visual fields from two NEI sponsored clinical trials on each disorder, ONTT and IIHTT, will be investigated to determine if the findings parallel the reported outcomes and effects of therapy. We will also test whether machine learning quantifiable archetypes, which are disease-associated patterns of field deficits, are similar to expert determinations, whether they are sensitive to changes in optic nerve function, and if they reveal residual optic nerve dysfunction in eyes reported to be normal by prior study criteria. Adding cases of IIH and optic neuritis from the clinic will enhance the archetypes for each disorder for use in the clinic and new studies.

Narrative Analysis of the visual fields of patients with optic neuropathies using machine learning will improve the evaluation and provide objective measurement, rather than the current descriptive methods. The approach, called archetypal analysis, should improve safety monitoring during clinical trials as well as uncover residual visual field deficits not seen with other types of analyses.