Artificial intelligence automation of urine culture reading with BD Kiestra Urine Culture Application: measurement of performance and potential efficiency gains.

Automation of microbiology sample inoculation and incubation has recently been shown to improve microbiological and clinical endpoints (such as turnaround time) as well as improve efficiency. However, instruments for automation of microbiology sample culture reading and interpretation have only recently become available for clinical evaluation. We evaluated the BD Kiestra Urine Culture Application (UCA) for culture interpretation of clinical urine samples and compared the results to the reading of Kiestra image by scientific staff. Of the 1021 urine samples processed by both UCA and scientist culture reading, 1003 (98%) yielded concordant results at 18-h incubation. The number of samples with ≥104 colony-forming units (CFU)/mL (≥107 CFU/L) which were correctly recognised by the UCA using the early growth detection algorithm was 424/424 (100%). We found the UCA to be an accurate artificial intelligence solution, and we describe the potential for large workload efficiency gains in addition to more rapid report turnaround times.