The use of digital image analysis in the histological assessment of Sjögren's syndrome salivary glands improves inter-rater agreement and facilitates multicentre data harmonisation

• Published in: Clinical and experimental rheumatology 2020-07, Vol.38 Suppl 126 (4), p.180-188
• Main Authors: Lucchesi, Davide, Pontarini, Elena, Donati, Valentina, Mandarano, Martina, Sidoni, Angelo, Bartoloni, Elena, Baldini, Chiara, Tappuni, Anwar R, Bombardieri, Michele
• Format: Article
• Language: English
• Subjects: Algorithms; Humans; Observer Variation; Reproducibility of Results; Salivary Glands - diagnostic imaging; Sjogren's Syndrome - diagnosis
• ISSN: 0392-856X

To assess whether the use of digital image analysis (DIA) in primary Sjögren's syndrome (pSS) for the calculation of the total area of the salivary gland (SG), focus score (FS) and SG area occupied by the inflammatory infiltrate (area fraction, AF), was able to generate reproducible readings among different raters, reducing disagreement. Haematoxylin and Eosin digital slides from pSS and non-specific chronic sialadenitis (NSCS) patients were analysed blindly by 4 independent raters among 3 centres. Using an open-source software (QuPath) raters were asked to provide the total area of the gland i) using a grid-based method and ii) a software-based area-calculation tool, iii) the number of inflammatory foci and iv) the total area of the inflammatory infiltrate. Collected data was used to calculate the inter-rater agreement. For the calculation of the total SG area, DIA generated higher agreement among raters than grid-based calculation (inter-class correlation coefficient ICC=0.85 vs 0.98). Agreement for calculated total area of the inflammatory infiltrate (ICC=0.94) and for AF (ICC=0.94) was higher than infiltrates count number (ICC=0.54) and FS (ICC=0.56). AF achieved a 30% improvement over the FS at generating consensus among raters when used as a diagnostic cut-off. A digital approach achieved a far superior inter-rater agreement when calculating the total area compared to a grid-based approach. The calculation of AF proved superior to FS in correctly classifying pSS vs NSCS biopsies. We suggest that digitally calculated AF should be used alongside FS for large multi-centre studies to improve data harmonisation.