Comparison of manual with artificial intelligence-aided interpretation of ANA HEp-2 IIF assay patterns in a clinical diagnostics lab

OBJECTIVES: Detection of antinuclear antibody (ANA) via indirect immunofluorescence (IIF) on HEp-2 cells is a screening test for the serological diagnosis of systemic autoimmune rheumatic diseases. Automated interpretation of ANA classification by novel artificial intelligence (AI)-aided pattern recognition was compared with expert reading under routine conditions. METHODS: Consecutive serum samples of 2671 individuals referred to a routine laboratory were analysed for ANA titers and patterns using the automated interpretation system akironNeo. AI-based ANA detection was compared with independent classification by two experienced immunologists according to the international consensus on ANA patterns (ICAP) competence level. RESULTS: Overall, a good agreement (κ > 0.60) between the different evaluators both for positive/negative classification of ANA fluorescence images as well as for the pattern classification of positive samples with a titer ≥ 1:320 was observed. Positive/negative differentiation at different cut-offs revealed κ values from 0.584 to 0.760 whereas corresponding pattern recognition for interphase, metaphase and cytoplasmic patterns demonstrated κ values from 0.560 to 0.736 for samples scored as positive by all three evaluators. CONCLUSIONS: The AI-based software showed a similar performance compared to human observers. AI-aided ANA image analysis can facilitate the diagnostic workflow of ANA IIF assays and reduce subjectivity during image classification.