An improved artifact removal algorithm for continuous cardiac output and blood pressure recordings

Measurement artifacts are common in hemodynamic recordings such as cardiac output and blood pressure. Manual artifact removal is cumbersome for large datasets, and automatic processing using algorithms may reduce workload and provide more reproducible outcomes. This paper presents an artifact removal algorithm which is more aggressive compared to a previously described method. The algorithm was evaluated on cardiac output (CO) and mean arterial pressure (MAP) recordings from 23 subjects measured by two different devices (LiDCO and Nexfin), and compared to the previously described method as a reference. Precision, recall and F-score was determined by agreement with manual inspection by an expert. Based on the total of all measurements from CO and MAP by LiDCO and CO and MAP by Nexfin, precision was 86%, 79%, 79% and 68% respectively (87%, 62%, 76% and 58% for the reference method), recall was 97%, 94%, 89% and 97% (31%, 6%, 28% and 6% for reference), F-score was 91%, 85%, 84% and 80% (46%, 10%, 41% and 10% for reference). The proposed algorithm offers an improved performance in removing true artifacts, in some cases a reduced ability to preserve true measurements, but an improved overall accuracy.