Abstract
Machine-learning (ML) models in flow cytometry have the potential to reduceerror rates, increase reproducibility, and boost the efficiency of clinicallabs. While numerous ML models for flow cytometry data have been proposed, fewstudies have described the clinical deployment of such models. Realizing thepotential gains of ML models in clinical labs requires not only an accuratemodel, but infrastructure for automated inference, error detection, analyticsand monitoring, and structured data extraction. Here, we describe an ML modelfor detection of Acute Myeloid Leukemia (AML), along with the infrastructuresupporting clinical implementation. Our infrastructure leverages the resilienceand scalability of the cloud for model inference, a Kubernetes-based workflowsystem that provides model reproducibility and resource management, and asystem for extracting structured diagnoses from full-text reports. We alsodescribe our model monitoring and visualization platform, an essential elementfor ensuring continued model accuracy. Finally, we present a post-deploymentanalysis of impacts on turn-around time and compare production accuracy to theoriginal validation statistics.