SepsisAI-Orchestrator: Open-Source Platform for Deploying ML Models in Hospitals
A team of researchers has introduced SepsisAI-Orchestrator, an open-source modular framework aimed at connecting machine learning innovations with practical applications for the early detection of sepsis. This platform features Clinical Document Architecture preprocessing inspired by HL7 FHIR, NoSQL data storage, a containerized LightGBM classifier accessible through REST APIs, and a Streamlit clinical dashboard, all managed using Docker and Kubernetes. The existing LightGBM model, validated with an F1 score of 0.87-0.94 on PhysioNet 2019, remains unchanged; the focus is on enhancing the infrastructure and evaluating its performance under stress. This initiative tackles systemic challenges like varied data formats, lack of standardized deployment processes, and discrepancies between research models and hospital demands. The findings are published in arXiv:2605.22331.
Key facts
- SepsisAI-Orchestrator is an open-source modular platform for deploying AI models in early sepsis detection.
- It integrates HL7 FHIR-inspired CDA preprocessing, NoSQL storage, a containerized LightGBM classifier via REST APIs, and a Streamlit dashboard.
- Orchestration uses Docker and Kubernetes.
- The LightGBM model achieves F1 0.87-0.94 on PhysioNet 2019 and is reused without modification.
- The platform addresses heterogeneous data, lack of standardized workflows, and latency requirements.
- Published on arXiv with ID 2605.22331.
Entities
Institutions
- arXiv