A multi-analyte simulation benchmark showed that machine-learning anomaly detection in clinical biochemistry laboratories did not reliably outperform traditional quality-control rules for common failures. The benchmark covered 54,000 quality-control sequences, 12 analytes, 5 scenarios, and 300 repetitions, with 500 runs per sequence. At a unified false-alarm operating point, the advantage of the two machine-learning model types over traditional methods such as the Westgard multirule narrowed, with the results pointing toward complementary multilevel monitoring.
Source:BIOENGINEER.ORG · Read Original
Leave a Reply