Diagnostic performance of urine analysis based on flow microimaging and artificial intelligence recognition technology in suspected urinary tract infection patients

Abstract

To investigate the diagnostic performance and clinical value of flow microimaging and artificial intelligence recognition technology for rapid screening of suspected urinary tract infection (UTI). Standard strains of bacteria responsible for UTIs, Escherichia coli ATCC25922 and Staphylococcus aureus ATCC25923 were prepared and used to evaluate the accuracy of classifying and counting bacteria. A total of 146 specimens of clean, midstream urine were collected from adults with suspected UTI (excluding pregnant women, and patients with urethral catheterization) and analyzed by urinary culture and assay using a MUS-3600 analyzer. Fourfold tables were used to evaluate the diagnostic performance in measurements of nitrite and leukocyte esterase. ROC curves were generated to identify cutoff values for bacillus, coccus, leukocyte, and leukocyte cluster counts. Of the samples cultured, 85 (58%) were negative and 61 (42%) were positive. E. coli and S. aureus showed good linear relationships between measured and theoretical values in a series of standard strains samples (R 2>0.95). The sensitivity of the nitrite parameter for diagnosis of suspected UTI was 37.7% and the specificity was 100%. Cutoff values obtained from ROC analysis were 50 μl for bacillus (sensitivity: 69.5%; specificity: 96.5%) with positive predictive value of 93.2% and negative predictive value of 82%. WBC ≥ 23 μl or bacillus ≥50 μl gave the highest sensitivity of 81% and NPV of 86%. MUS-3600 demonstrated sufficient specificity for UTIs to have utility in reducing needless urinary culture. Accuracy in classifying bacillus shows great potential for the rapid identification of pathogens for clinical purposes.