Stimulated emission depletion (STED) microscopy for high-resolution ANA pattern imaging and machine learning comparison with confocal microscopy

Antinuclear antibodies (ANA) are established as biomarkers for the diagnosis and monitoring of autoimmune diseases. Their detection and pattern differentiation rely on indirect immunofluorescence (IIF) on human epithelial type 2 (HEp-2) cells, with fluorescence patterns visually assessed by diagnosticians. However, these approaches have inherent limitations, including dependence on observer experience and limited resolution. To address these limitations, higher-resolution imaging and automated analysis are needed to better capture and interpret ANA pattern complexity. In this study, we developed and compared machine learning models for ANA pattern classification based on images acquired for the first time using super-resolution stimulated emission depletion (STED) microscopy and conventional confocal microscopy. The dataset comprised 2920 confocal and 2907 STED images, cropped to single nuclei and balanced through data augmentation. Maximal model accuracy on confocal images reached 94.8%, whereas accuracy for models trained on STED images was lower, reaching 87.8%. While STED provides higher-resolution imaging of ANA patterns, classification performance was inferior to confocal images, likely reflecting a misalignment with current ANA pattern classification and the increased level of detail captured by STED imaging. Our dataset represents the first collection of ANA images acquired by STED microscopy.