AI Analysis of ANA Images Predicts Selected Autoantibodies in Temporal Validation

Researchers evaluated whether artificial intelligence could predict specific autoantibodies from human epithelial type 2 antinuclear antibody (ANA) indirect immunofluorescence images. The development cohort included 47,890 eligible patients from 512,959 assessed; mean age was 55 years, and 78% were female. Eligibility required an ANA titer ≥1:80 and specific autoantibody testing within 90 days. Two AI architectures were trained per antibody using the first ANA test per patient from December 2016 through December 2022.

Models for centromere, double-stranded DNA (dsDNA), Smith (Sm), SSA, and SSB antibodies achieved areas under the receiver operating characteristic curve (AUROCs) ≥0.80. Centromere prediction performed best, with AUROC 0.95, accuracy 94.2%, and specificity 95.3%. Compared with associated immunofluorescence patterns, dsDNA and Sm models more frequently correctly classified antibody-negative cases: 88.2% versus 54.7% and 82.6% versus 60.3%, respectively. Classification of antibody-positive cases was similar. AUROCs were lower for RNP (0.68) and Scl-70 (0.59).

Prospective temporal validation in 21,427 patients tested from January 2023 through August 2024 showed similar performance. DsDNA discrimination remained stable across an assay transition, with AUROC approximately 0.85. The findings suggest potential for standardizing ANA interpretation and guiding confirmatory testing. However, the population was restricted to ANA-positive patients, comparisons used simplified pattern-based approaches, and the abstract does not establish whether AI-guided testing improves clinical outcomes or safely reduces confirmatory testing.