Beyond structural assessment, OCT is increasingly used to monitor treatment response and predict ON relapse. Notably, RNFL thinning > 20 μm combined with prolonged visual evoked potential (VEP) latency > 12 ms significantly increased relapse risk (hazard ratio = 2.7, p = 0.003) (24). Additionally, early RNFL thinning correlated with reduced retinal microvascular density, and their combination enhanced relapse prediction accuracy (25). Building on structural-functional correlations, a predictive model based on OCT-derived GCIPL metrics in traumatic optic neuropathy demonstrated that GCIPL thickness served as an independent prognostic factor for visual recovery, achieving an area under the curve (AUC) of 0.90 within 1 year (26). Further study confirmed that accelerated longitudinal GCIPL thinning over 6–12 months is strongly associated with poorer visual outcomes and increased relapse risk (27). OCT also plays a role in monitoring disease progression within the framework of No Evidence of Disease Activity (NEDA) in MS, where preserved retinal thickness correlates with sustained NEDA status and reduced relapse risk (28).

Recent studies also provide disease-specific insights for MOGAD and NMOSD, highlighting structural changes along the anterior and posterior visual pathways that reflect distinct neurodegenerative patterns. OCT provides valuable information not only for prognosis but also for differentiating ON subtypes. GCIPL thinning, particularly in the parafoveal region, is more pronounced in NMOSD-associated optic neuritis (NMOSD-ON) than in multiple sclerosis–associated optic neuritis (MS-ON), and strongly correlates with cognitive and visual processing speed deficits (r = −0.62, p < 0.001) (23). Further studies have demonstrated that MS-ON primarily affects the temporal RNFL quadrants, AQP4-IgG + NMOSD-ON results in diffuse loss across all quadrants, and MOGAD-ON exhibits relatively symmetric but milder thinning (p < 0.001) (29).

In MOGAD, eyes with prior ON show significant reduction in pRNFL and GCIPL thickness, which correlates with the number of ON episodes, as well as structural atrophy along the visual pathway, including lateral geniculate nucleus (LGN) and occipital cortex, indicating pathway-specific neurodegeneration (30). In NMOSD, LGN volume is reduced after ON, associated with retinal neuroaxonal loss and optic radiation damage, but does not decline in the absence of new ON episodes, suggesting that neurodegeneration primarily follows acute attacks (22). Collectively, these findings highlight OCT’s value in capturing disease-specific progression patterns across different ON etiologies.

In line with these findings, it has been observed that despite severe acute vision loss, MOGAD-ON often retains relatively preserved GCIPL thickness (mean reduction 14 μm), in contrast to NMOSD-ON eyes, which show markedly greater loss (mean reduction 28 μm, p = 0.008) (31). Recent comparative reviews and cohort analyses provide consistent evidence that MOGAD-ON is frequently bilateral at onset, associated with pronounced optic disc swelling, and followed by variable but sometimes modest chronic GCIPL loss despite profound acute deficits. By contrast, AQP4-IgG + NMOSD-ON is characterized by severe and diffuse chronic pRNFL and GCIPL thinning after attacks, with average chronic pRNFL values often <80 μm, whereas MS-ON more commonly shows temporal-predominant RNFL loss (19). These modality-specific signatures support the use of OCT-guided triage for serologic testing when clinical features are ambiguous (Table 1).

Beyond descriptive features, newer analytic models further enhance diagnostic classification. An OCT-based model was developed that accurately distinguished diffuse RNFL and GCIPL thinning in NMOSD-ON, temporal thinning in MS-ON, and relatively preserved retinal structure in MOGAD-ON despite significant visual loss (32). Similarly, inter-eye difference (IED) metrics were shown to be useful, with GCIPL IED significantly greater in MOGAD-ON compared to MS-ON and NMOSD-ON, achieving a sensitivity of 87% and specificity of 85% for subtype discrimination (33). Finally, longitudinal MS cohorts demonstrate the prognostic dimension of OCT biomarkers, with pRNFL quantification identifying individuals at increased risk of cognitive dysfunction (23). Together, these findings underscore the capacity of OCT-derived metrics to capture both disease-specific phenotypes and long-term outcomes more comprehensively.

High-speed swept-source OCT (SS-OCT) has been shown to significantly enhance imaging depth and speed, improving visualization of deep optic nerve structures and reducing motion artifacts, thereby enabling more accurate detection of optic disc edema and structural distortion in ON patients (34). Quantitative birefringence analysis using polarization-sensitive OCT (PS-OCT) has been reviewed, highlighting its ability to detect pre-atrophic microstructural disorganization in the RNFL, which may aid in identifying ON activity before irreversible damage occurs (35). A novel search algorithm for OCT layer segmentation has been developed, achieving highly precise delineation of retinal layers even in severely distorted optic nerves, thereby facilitating robust ON subtype classification and treatment monitoring (36). Looking ahead, the integration of advanced OCT modalities with artificial intelligence and longitudinal data analysis is expected to further enhance the early detection, classification, and individualized monitoring of ON, paving the way toward predictive and precision neuro-ophthalmology.

OCTA has revealed significant reductions in peripapillary and macular vessel densities during both acute and chronic stages of ON, offering valuable insights beyond traditional structural metrics. Quantitative studies of demyelinating ON have shown that vessel density loss in the superior and inferior quadrants is significantly more pronounced (p < 0.001) and closely correlates with visual field defects (r = 0.72, p < 0.01) (37). Additionally, peripapillary choroidal microvasculature dropout has been associated with poor visual recovery in ON patients (p = 0.003), even after adjusting for RNFL thickness, suggesting that deeper vascular alterations may serve as independent prognostic biomarkers (38). Early OCTA-detected reductions in vessel density have been shown to precede structural thinning and predict long-term functional decline, particularly in patients experiencing recurrent ON episodes (25). Together, these results highlight OCTA’s potential as a sensitive, non-invasive modality for early detection, prognostication, and treatment response evaluation in ON.

A systematic review of OCTA biomarkers in MS and NMOSD identified consistent patterns of pronounced superficial capillary plexus (SCP) rarefaction in NMOSD, with peripapillary vessel density reductions exceeding 20%, whereas MS exhibited milder changes primarily in the temporal quadrant (39). These findings reinforce the value of SCP integrity as a subtype-specific imaging biomarker. Mohammadi et al. (40) conducted a meta-analysis of OCTA data across 11 studies, reporting that NMOSD patients had significantly larger foveal avascular zone (FAZ) areas and lower parafoveal vessel densities than both MS and healthy controls (p < 0.001), while MOGAD patients exhibited near-normal microvascular metrics. This highlights the utility of FAZ morphology and flow indices in distinguishing NMOSD-ON from other ON subtypes. Furthermore, novel OCTA-derived indices, including vessel tortuosity and fractal dimension, have been introduced to evaluate microvascular complexity (39). Findings indicate that NMOSD-ON is associated with markedly reduced capillary regularity and impaired vascular remodeling potential, providing mechanistic insights and supporting OCTA’s role in disease classification and activity monitoring.

Recent innovations in deep learning (DL) have significantly enhanced the accuracy, efficiency, and reproducibility of OCTA analysis. Several state-of-the-art DL frameworks have been reviewed, highlighting the robustness of U-Net variants and attention-based models in accurately segmenting fine vascular structures, achieving vessel segmentation Dice coefficients exceeding 0.90 and reducing manual annotation by over 70%, thereby paving the way for standardized and fully automated ON-related vascular imaging analysis (41). A multi-view tri-alignment deep learning framework, MuTri, has been proposed to translate structural OCT into synthesized OCTA volumes with high fidelity, overcoming inter-modality variability. This model achieved over 25% improvement in cross-modality translation accuracy compared to baseline GAN architectures, enabling simulated 3D OCTA generation in cases where perfusion data may be unavailable (42). Collectively, these developments position DL as a critical tool for earlier and more precise characterization of microvascular changes in ON and related demyelinating conditions.

Conventional MRI sequences, including T2-weighted, STIR, and contrast-enhanced T1-weighted imaging, remain foundational for detecting optic nerve lesions, enhancement, and demyelination in ON and related disorders, providing essential diagnostic and prognostic information (43). Radiological predictors of visual outcome in ON have been investigated, showing that greater optic nerve lesion length and higher enhancement intensity during acute episodes are significantly associated with poorer visual recovery at 12 months (r = −0.71, p < 0.001), highlighting inflammatory load as a key prognostic indicator (44). Furthermore, contrast-enhanced MRI studies have shown that optic nerve enhancement lengths exceeding 17 mm are significantly correlated with initial deficits in high-contrast visual acuity and contrast sensitivity (p < 0.05), reinforcing the utility of MRI enhancement metrics in gauging acute disease severity and guiding early treatment decisions (45).

Although optic nerve MRI is currently optional for MS diagnosis, it provides crucial information for lesion assessment, monitoring disease progression, and predicting visual outcomes (46). Thus, conventional MRI sequences, together with emerging quantitative assessments, serve both diagnostic and prognostic roles, enhancing ON clinical evaluation.

DTI has emerged as a sensitive imaging modality to capture microstructural changes in the optic nerve. The use of DTI in optic neuropathies has been reviewed, highlighting that elevated apparent diffusion coefficient (ADC) and reduced fractional anisotropy (FA) in the acute phase are predictive of subsequent axonal degeneration and vision loss, supporting early DTI assessment as a reliable predictor of long-term structural damage (47). The pathophysiological mechanisms underlying optic neuritis in multiple sclerosis have been reviewed, highlighting that DTI metrics reliably reflect microstructural axonal damage and demyelination processes in the optic nerve, thus serving as essential biomarkers for disease progression (48). Quantitative spinal MRI studies have demonstrated that structural changes in the spinal cord following optic neuritis closely relate to optic nerve damage and visual impairment, suggesting broader CNS involvement detectable by advanced imaging (49). Overall, these findings support DTI as a sensitive tool for capturing microstructural changes predictive of visual outcomes in ON, offering valuable insights for tailored therapeutic interventions.

Optic nerve MRI findings have been shown to correlate with cerebral MRI changes in MS, supporting the idea that both conventional and non-conventional imaging can provide a comprehensive view of disease activity (50). The value of DTI and other quantitative MRI sequences for assessing optic nerve microstructure has been emphasized, highlighting their potential to predict long-term visual outcomes and guide early therapeutic interventions (46).

MRI features can aid in distinguishing ON etiologies. A retrospective analysis of 56 ON patients showed that MOGAD-ON more frequently presented with bilateral involvement and lesions extending to distal segments, such as the intraorbital and canalicular regions, which was distinct from MS-ON and NMO-ON (p = 0.006 and p = 0.039, respectively) (51). Furthermore, combining brain/spinal cord and optic nerve MRI features with OCT RNFL thickness enabled differentiation of MS from NMOSD/MOGAD with 95% classification accuracy (p < 0.001) (52).

MRI patterns, including optic nerve vs. sheath involvement, can distinguish typical (MS-related) from atypical ON (NMOSD- or MOG-IgG-related), guiding tailored treatment strategies (53). In addition, MRI features of intraocular optic nerve disorders have been summarized, highlighting the role of MRI in lesion characterization, extent assessment, and monitoring therapeutic response (54). These findings collectively highlight that MRI, when combined with advanced sequences and OCT, improves etiological classification and informs individualized management in ON.

Comparative analysis of DTI parameters in white matter tracts of MS and related disorders demonstrated that decreased axial diffusivity and fractional anisotropy strongly correlate with retinal nerve fiber layer thinning and visual evoked potential abnormalities, reinforcing DTI’s utility for early detection and longitudinal monitoring of microstructural injury (55).

Integration of OCT and MRI enhances understanding and management of ON by linking structural and functional changes. Enhanced depth imaging OCT was used to identify retrolaminar hyper-reflective foci in MS patients with acute ON, revealing significant associations with MRI-detected lesions (p = 0.000) (56). Moreover, OCT-measured pRNFL thinning in 50 patients with optic neuropathy, specifically linked to ON, was shown to complement MRI assessments, with significant correlations to lesion length (p = 0.01), indicating that OCT enhances MRI’s diagnostic precision by capturing retinal changes associated with optic nerve damage (57). This approach was further validated in 79 MS patients with ON, showing that incorporating OCT-assessed optic nerve data into MRI-based dissemination in space (DIS) criteria increased diagnostic sensitivity by 12% (p = 0.03), improving early ON diagnosis without compromising specificity, thereby highlighting the synergistic potential of these modalities (58). Overall, these findings underscore the synergistic potential of combining structural and functional biomarkers, linking retinal and optic nerve pathology to enhance both diagnostic accuracy and therapeutic decision-making, paving the way for a more holistic understanding of ON pathophysiology.

A convolutional neural network (CNN) model was employed to infer superficial capillary blood flow maps from standard OCT images, achieving 85% prediction accuracy and a Dice coefficient of 0.82 across independent ON samples (59). This demonstrated that microvascular perfusion could be assessed without OCTA. Cassottana et al. (60) conducted a comparative study assessing papillary and macular blood flow using OCTA in healthy subjects and patients with various optic neuropathies, providing normative and pathological benchmarks that enhance the interpretation of inferred flow maps. Together, these approaches strengthen deep learning models’ ability to discriminate ON etiologies by integrating structural and vascular imaging data.

An enhanced OCTA post-processing pipeline using attention-guided super-resolution algorithms was introduced, significantly improving vascular detail resolution in retinal imaging, with PSNR increased by 3.9 dB and SSIM by 0.06 compared to baseline interpolation methods (61). This method enables more accurate visualization of capillary-level changes in optic neuropathies, facilitating earlier and finer diagnostic discrimination. Moreover, a multimodal deep learning framework integrating structural OCT, OCTA, and serum omics data was developed, achieving an AUC of 0.95 for distinguishing inflammatory versus ischemic optic neuropathies and predicting six-month visual outcomes with 91.2% sensitivity and 89.5% specificity (62). These advances underscore the value of multimodal fusion in enhancing ON diagnostic precision and outcome prediction.

High-resolution 3D OCT volumes combined with synthetic slice generation were leveraged to improve lesion detection in demyelinating diseases, achieving 92.3% diagnostic accuracy in distinguishing MS-related optic neuropathy from controls and enhancing visualization of peripapillary microstructural abnormalities (63). In parallel, Vision Transformer (ViT) technology was applied for 3D reconstruction of OCTA images, resulting in a 20% improvement in accuracy (64), enabling more comprehensive depiction of tissue and blood flow changes in various ON types. Additionally, AI groups have optimized segmentation strategies via reinforcement learning, reducing noise error by 15% and enhancing microvascular change detection across ON etiologies (41). Collectively, these developments position DL as a critical tool for more precise characterization of microvascular changes and overall ON imaging.