Leber’s Hereditary Optic Neuropathy visual field loss characterization by machine learning

Abstract

Purpose : Characterization of chronic Leber’s Hereditary Optic Neuropathy (LHON) visual field loss patterns through unsupervised machine learning

Methods : LHON patients in the chronic phase with visual field (VF) tests performed by SITA standard 30-2 or 24-2 Humphrey VF analyzer (Carl Zeiss Meditec, Dublin, CA, USA) were considered. To ensure that the dataset would be made up of chronic stable stage patients, VFs performed in the first two years after disease onset were excluded. Patients with and without visual acuity (VA) recovery were enrolled, defining VA recovery as a change of more than 0.2 decimal or from off-chart to on-chart vision after initial loss of vision. For each eye, two VFs were collected to enlarge the dataset. Based on the VF’s total deviations (TDs) and resorting to Python 3.8 (Anaconda), an Archetypal Analysis (AA) model is being developed for the identification and quantification of the patterns of VFs loss underlying this type of patient.

Results : Hundred ninety-eight VF tests were collected from 155 eyes of 78 chronic LHON patients. From the preliminary implementation of AA for the standardized analysis of regional VF loss patterns in LHON patients, 12 archetypes (AT) were identified. The ATs describing defects, included central scotoma defects with varying severities, namely a more localized central scotoma which is commonly seen for patients with VA recovery, or a more enlarged central scotoma expected for cases without VA recovery. Also, an AT resembling VF fenestration incorporated the AA-model.

Conclusions : Identification of visual loss patterns of LHON patients in a chronic phase poses a challenge, as well as the distinction and association of patterns to patients that recover or not vision. Therefore, the proposed application of unsupervised machine learning to VFs might be of support to standardize the VF analysis in these types of patients, enabling the distinction and quantification of visual loss patterns of a VF of a patient. Furthermore, characterization of patients with or without visual recovery could provide insights for both clinical and gene therapies.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.