Accuracy of classification of urinary Gram-stain findings by a computer-aided diagnosis app compared with microbiology specialists

Introduction. Timely and accurate diagnosis of bacterial infections enables early administration of appropriate antimicrobial treatment and improved outcomes. Hypothesis/Gap Statement. The accuracy of computer-aided diagnosis (CAD) for identifying organisms on urine Gram stains has not been compared with that of microbiology specialists (MS). Aim. To compare the interpretation of urine Gram-stain results by MS and a CAD app designed using artificial intelligence. Methodology. Urine specimens from patients with urinary tract infections were used and collected at two tertiary hospitals between 1 April and 31 December 2022. Using non-inferiority analysis to assess whether CAD was non-inferior to expert interpretation, CAD-predicted microscopic findings of the Gram-stained slide generated from iPhone camera images from two hospitals were compared with those from ten MS. A total of 153 images were taken from each hospital, and CAD interpreted a total of 306. The primary endpoint was the prediction accuracy based on the morphology of the Gram-stained bacteria. Results. The accuracy (95% confidence interval) of MS and CAD predictions was 83.0% (81.6%–84.3%) and 87.9% (83.7%–91.3%), respectively, with a difference of –4.93% (–8.43% to –0.62%) indicating non-inferiority of CAD. Conclusion. CAD was non-inferior to MS predictions for identifying Gram-stained pathogens; therefore, CAD was suggested to have the potential for guiding empirical antibiotic selection in patients with urinary tract infections.