Development and Temporal Validation of Artificial Intelligence (AI) ‐Enhanced Human Epithelial Type 2 Indirect Immunofluorescence Image Analysis for Specific Autoantibody Prediction in Systemic Autoimmune Rheumatic Diseases
Authors' conclusion
AI applied to ANA IIF images identified specific autoantibodies and often improved classification of antibody-negative cases versus simplified pattern-based comparators.
Study context
- Population
- Antinuclear antibody (ANA)-positive patients (human epithelial type 2 indirect immunofluorescence (IIF) titer ≥1:80) with anti-dsDNA, anti-Sm, ribonucleoprotein (RNP), SSA, SSB, Scl-70 or centromere antibody (CMA) testing within 90 daysmean age 55, 78% female
- Participants
- 47,890 patients
- Comparison
- AI applied to ANA IIF images vs IIF patterns for specific autoantibody prediction
- Outcome
- AUROC, sensitivity, specificity and accuracy for each autoantibody
Methodological detailsdevelopment and temporal validation, prospective validation cohort
- Development and temporal validation
- Prospective validation cohort
Performance
As reported in the abstract
Discrimination of specific autoantibodies by AI models (development cohort)
| Test or model | AUC | Accuracy | Specificity |
|---|---|---|---|
| CMA model | 0.95 | 94.2% | 95.3% |
CMA model
AUC0.95
Accuracy94.2%
Specificity95.3%
Further findings
Antibody-negative cases classified as negative, vs IIF patterns
dsDNA
Sm
Antibody-negative cases classified as negative, vs IIF patterns
- Similar classification of antibody-positive cases
Temporal validation (prospective cohort, n = 21,427)
- dsDNA discrimination stable across assays (AUROC)
- ≈0.85
- Performance similar
Parts of this Brief are withheld: the automatic checks could not confirm them against the abstract.
Original publication
Development and Temporal Validation of AI ‐Enhanced Human Epithelial Type 2 Indirect Immunofluorescence Image Analysis for Specific Autoantibody Prediction in Systemic Autoimmune Rheumatic Diseases. Arthritis Rheumatol. Published 28 August 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →