The impact of OSCAR-IB quality control and manual segmentation on retinal atrophy rates in multiple sclerosis

Inner retinal layer thinning, particularly of the peripapillary retinal nerve fiber layer (pRNFL) and the ganglion cell-inner plexiform layer (GCIPL), as measured by optical coherence tomography (OCT), is an established surrogate biomarker of neuroaxonal damage in multiple sclerosis; however, the influence of preprocessing strategies on longitudinal retinal change estimates remains unclear. We investigated the effects of OSCAR-IB quality control (QC) and manual segmentation correction on retinal layer change rates.

People with MS (pwMS) with ≥ 2 OCT routine scans obtained ≥ 6 months apart were included. Annualized pRNFL and GCIPL thickness change was estimated using linear mixed-effects models across four preprocessing strategies: raw data, data after full OSCAR-IB QC, data after QC excluding the algorithm-criterion for segmentation, and data after manual segmentation correction. Longitudinal stability was evaluated using residual variance and random slope variance.

A total of 173 pwMS (mean age 34.6 years [8.5], 72.3% female) were included. Mean annualized pRNFL and GCIPL thickness change in raw data were -0.29%/year (0.29) and 0.14%/year (0.37), respectively. Compared with raw data, application of OSCAR-IB QC was associated with lower residual variance for both pRNFL (1.50 vs. 1.73) and GCIPL (0.65 vs. 0.78). Manual segmentation correction did not improve longitudinal stability for either retinal layer.

OSCAR-IB QC was associated with improved longitudinal consistency of pRNFL and GCIPL change estimates in MS. In contrast, manual segmentation correction was not associated with additional benefit beyond exclusion-based QC, suggesting limited added value in routine clinical or large-scale research settings where feasibility and scalability are critical.