Invest Ophthalmol Vis Sci. 2026 Jul 1;67(8):8. doi: 10.1167/iovs.67.8.8.
ABSTRACT
PURPOSE: To use the machine learning algorithm archetypal analysis (AA) to characterize visual field (VF) loss patterns in patients with chronic Leber hereditary optic neuropathy (LHON) and to determine whether these VF patterns differentiate for patients who recovered visual acuity (VA).
METHODS: For patients with molecularly confirmed LHON, we retrospectively collected 30-2 or 24-2 VFs (Humphrey VF analyzer) performed at least 3 years after disease onset. A total of 220 VFs (220 eyes, 117 patients) were used as input for the AA algorithm. Archetypes (ATs) and relative weights were statistically compared between VA-recovered and not-recovered groups and among primary mitochondrial DNA mutations. K-means clustering and principal component analysis enabled two-dimensional visualization of the relationship among VFs, ATs, mean deviation, and VA.
RESULTS: The LHON-AA model consisted of seven ATs: AT1 resembled a total loss VF pattern (30%) and AT2 a normal VF (23%). Small (AT3, 15%) and large (AT4, 10%) central scotomas followed as representative patterns. AT1 was significantly more prevalent in the not-recovered group, whereas AT2 and AT3 predominated in the recovered VA group. Less severe patterns were more present for m.14484T>C/ND6 than the other primary mutations. Two-dimensional visualization highlighted the nonlinear relationship between VA and VF outcome.
CONCLUSIONS: AA is a robust method for categorizing and quantifying VF damage in patients with chronic LHON, providing novel insights into the complexity of the disease. The findings suggest that the recovery of VA and VF damage are not transposable, although not entirely independent either, and that both parameters should be considered in the comprehensive assessment of functional recovery in LHON.
PMID:42390170 | DOI:10.1167/iovs.67.8.8