Background: Nailfold videocapillaroscopy (NVC) is a fundamental tool for the early diagnosis of systemic sclerosis (SSc) and for differentiating Raynaud’s phenomenon (RP). However, its widespread clinical application is often hindered by inter-observer subjectivity and the time-consuming nature of manual assessment. Recent validation studies of AI-based algorithms, such as CAPI-Score and CAPI-Detect, have demonstrated high accuracy in identifying disease patterns[1,2].

Objectives: To evaluate the diagnostic performance and clinical utility of an AI-driven automated software in discriminating scleroderma from non-scleroderma patterns within a clinical cohort of patients with connective tissue diseases (CTD).

Methods: The study included patients with SSc, systemic lupus erythemathosus, idiopathic inflammatory myopathies, overlap syndromes and Raynaud’s syndrome complying with the respective EULAR diagnostic and classification criteria. Capillaroscopic examination was performed using the Smart G-Scope™ Auto Focus Digital Microscope & Telescope capillaroscope. NVC images were captured and analyzed using the AI-based software (Capillary.io). The AI software’s output was validated against the consensus of a certified expert capillaroscopist with more than 5 years of experience in capillaroscopic diagnostic (gold standard). Diagnostic metrics, including sensitivity, specificity, accuracy, ROC analysis and regression analysis were calculated.

Results: This prospective study included 88 patients (83% female and 17% male, mean age 52.5 ± 14.2 years) diagnosed with SSc (limited SSc: 15.9%, diffuse SSc: 12.5%), SLE (27.3%), primary RP (43.2%) and overlap syndromes (1.1%). The AI software achieved a sensitivity of 97.2%, specificity of 100.0%, and an overall accuracy of 98.9%. The inter-rater reliability between the AI and the expert rater was nearly perfect (Cohen’s kappa = 0.976; p < 0.001). ROC analysis identified capillary density (AUC = 0.924, optimal cutoff 6.75) and percentage of giant capillaries (AUC = 0.961, optimal cutoff 1.2%) as the most robust predictors of SSc. Correlation analysis revealed strong negative associations between capillary density and the percentage of giant (r = -0.757) and enlarged capillaries (r = -0.599). Subgroup analysis showed that anti-Scl70 antibodies were significantly associated with lower capillary density (p=0.016) and higher percentages of giant capillaries (p=0.019). No significant impact of glucocorticoids or most immunosuppressants was found on capillary metrics in the Scleroderma pattern group, with the exception of methotrexate, which was marginally negatively associated with capillary density (p=0.045).

Conclusions: Preliminary data suggest that AI-driven automated capillaroscopy provides exceptionally high diagnostic accuracy, potentially surpassing traditional manual assessment in clinical efficiency. It demonstrated excellent discriminatory ability between SSc and non-SSc patterns. Further evaluation of intra-rater variability and cost-effectiveness is required to fully validate its integration into routine rheumatological practice.

REFERENCES: [1] Gracia Tello BD, Sáez Comet L, Lledo G, Freire Dapena M, Mesa MA, Martín-Cascón M, Guillén del Castillo A, Martínez Robles E, Simeon-Aznar CP, Todolí Parra JA, Varela DC. Capi-score: a quantitative algorithm for identifying disease patterns in nailfold videocapillaroscopy. Rheumatology. 2024 Dec;63(12):3315-21.

[2] Lledó-Ibáñez GM, Sáez Comet L, Freire Dapena M, Mesa Navas M, Martín Cascón M, Guillén del Castillo A, Simeon CP, Martinez Robles E, Todolí Parra J, Varela DC, Maldonado G. CAPI-Detect: machine learning in capillaroscopy reveals new variables influencing diagnosis. Rheumatology. 2025 Jun;64(6):3667-75.

Acknowledgments: NIL.

Disclosure of Interests: None declared.