Artifact detection (AD) is a critical component of Clinical Decision Support Systems (CDSS) to assess the quality of data and issue clinical alarms with high specificity and low false alarm rates. Although many advanced AD algorithms have been developed, very few have been evaluated in a real-time environment or reached actual clinical implementation. Furthermore, no single AD algorithm is necessarily best-suited to serve the varying needs of a CDSS. This identifies the need to develop and evaluate a framework that supports the interoperability of a range of configurable AD algorithms that could be integrated with CDSS.
This thesis develops an AD framework to address six different shortcomings. This framework supports the following six features: (f.1) Flexibility to serve the needs of patient populations from different types of Critical Care Units (CCU) through generalization and customizability; (f.2) Reusability across multiple types of physiologic data harvested by different Original Equipment Manufacturer (OEM) monitors; (f.3) Standardized definitions of Signal Quality Indicators (SQI) that promote interoperability and comparison between independently developed components; (f.4) Reusability and scalability by mixing and matching several AD, Parameter Derivation (PD), and Clinical Event Detection (CED) components in various combinations; (f.5) Customizability to evaluate and compare the performance of multiple combinations of independently developed components on offline and potentially real-time patient data when integrated with clinical workflows; and (f.6) Standardized component interfaces that can potentially support real-time clinical implementation of AD.
The developed framework provides for a unique test bed with the ability to create new AD composition pipelines by mixing and matching independently developed or de-coupled AD components. The performance of these pipelines can be evaluated in integration with CDSS. A novel contribution of this research is a Common Reference Model (CRM) that defines standard interfaces to allow for interoperability of framework components. The CRM defines patient data attributes including data types and their SQI.
The framework is validated in a clinical SpO2 alarms generation study designed to evaluate four different AD composition pipelines. The results quantify and compare sensitivity and false alarm rates across the four experiments. This study demonstrates and validates the six framework features (f.1)-(f.6).