A certainty-weighted decision model for the detection of suction in VADs

Suction is a dangerous state of operation of rotary ventricular assist devices that must be detected and avoided. Schemes using physiologic indices have been proposed to detect this phenomenon. The paper describes a detection model, the certainty-weighted detection system (CWDS), that combines classification by indices to determine whether suction exists. The CWDS also provides an uncertain classification if indices are not sufficiently reliable to draw a definite conclusion. The CWDS is compared to detectors based on fuzzy and Dempster-Shafer theory that provide uncertain classifications. The assessment of performance of these detectors is based on a new measure of cost that quantifies the inclusion of uncertain classifications. The results show that all the structures have similar performance and that they may provide improvements over traditional schemes depending on the cost ratio between uncertain classifications and errors classifications