Diagnostic model of microvasculature and neurologic alterations in the retina and optic disc for lupus nephritis.

Background: Machine learning (ML) analysis of retinal nerve fiber layer (RNFL) thickness and vessel density (VD) alterations in the macular region and optic disc may provide a new diagnostic method for lupus nephritis (LN). This study aimed to assess these alterations in LN patients using optical coherence tomography angiography (OCTA).

Methods: A retrospective analysis was conducted on 81 systemic lupus erythematosus (SLE) patients without retinopathy, divided into two groups: LN (41 patients) and non-LN (39 patients). OCTA imaging was performed on all participants. Independent risk factors were identified through univariate and multivariate analyses, followed by the development of a random forest (RF) diagnostic model.

Results: A total of 37 RNFL and VD variables from the macular region and 23 from the optic disc were analyzed. Through elastic net regression, 16 significant factors were identified. Further multivariate logistic regression selected 8 critical factors, which were used to construct the RF model. The RF model achieved an area under the curve (AUC) of 0.950 (95 % CI: 0.882 to 1.000), accuracy of 0.903 (95 % CI: 0.743 to 0.980), sensitivity of 0.867, specificity of 0.938, a positive predictive value (PPV) of 0.929, and a negative predictive value (NPV) of 0.882.

Conclusion: This study highlights the potential of ML-based OCTA data in diagnosing LN. Key diagnostic factors included perimeter (PERIM), superficial capillary plexus vessel density (SVD) - parafoveal (para)-temporal (T), SVD-perifoveal (peri)-inferior (I), RNFL-Fovea, RNFL-Peri, RNFL-Peri-T, capillary-whole-image, and peripapillary RNFL (PRNFL)- inferonasal (IN).