Automation of a Portable Heart-Lung Machine and Patient Monitoring with Data Mining Methods

A cardiogenic shock is associated with a high mortality rate. Often, patients suffer from irreversible organ damage before they reach a hospital for professional therapy. In recent years miniaturized extracorporeal circulatory support systems increased the survivor rate by providing primary care for the patient at the emergency site and during transportation. Nevertheless, the limited space in ambulances, the lack of trained staff and financial considerations constrain an area-wide employment of such systems. An automatically regulated support system could provide optimal perfusion, increasing patient safety and reducing the workload of the emergency team. This work examines two aspects of the automation of a portable heart-lung machine: The design of robust controllers for the regulation of the heart-lung machine, and the application of Data Mining methods for intelligent patient monitoring. Animal experiments allowed to collect vital patient parameters and their interaction with the heart-lung machine in distinct scenarios. Based on this data a simulation model and a hydraulic circulatory model are established. The hydraulic model replicates the cardiovascular system and is used to design robust controllers, regulating the pump speed of the heart-lung machine dependent on blood pressure and flow. Four concurrent control strategies were implemented: a PI Controller, a H∞-Controller, a Fuzzy Controller and a Model Reference Adaptive Controller. The controllers are tuned robustly and evaluated in several scenarios. Autonomous circulatory support systems require continuous and failure-safe patientmonitoring. Usually, the medical practitioner decides on the choice of treatment, based on current patient parameters and his experience. Only in recent years Data Mining and Knowledge Discovery methods found their way into medical research. In this work Data Mining methods for the online assessment of the patient’s condition are reviewed. In a benchmark study it is shown how such algorithms are able to reduce the false alarm rate of patient monitors. This increases the quality of care and supports the application of autonomous medical devices such as the controlled heart-lung machine.