Advanced Suction Detection for an Axial Flow Pump

Abstract

An automatic detection system for ventricular collapse was developed and tested in a first clinical trial as part of a physiological speed control concept for axial flow pumps. From this clinical experience, and based on the acquired data during this trial, an optimization of the developed system was performed. An already-existing database of 784 individual cases was extended. For harmonization of this database an additional 412 snap files were extracted from continuous data recordings and classified manually using a standardized procedure. The already-developed and clinically tested algorithms were supplemented by one additional indicator derived from a preexisting criterion. One threshold value was replaced by application of a numerically optimized nonlinear characteristic curve dependent on heart rate. Finally, in a multidimensional optimization process of the entire suction detection system, 7 individual indicators were adjusted by using 17 independent threshold values. The optimization criteria were applied using a three-level hierarchical system. Within the final database consisting of 1196 snap shots the overall amount of maldetections could be reduced to 23 cases including 5 false positive events (0.42%) and 18 false negative decisions (1.5%). By application of the clinical experience from the first clinical trial of a physiologic control system it became possible to optimize the sensitivity and specificity of the suction detection system to unprecedented accuracy.