Reliability of B-line quantification by different-level observers and a software algorithm using point-of-care lung ultrasound.

Quantification of B-lines on lung ultrasonographs is operator-dependent and considered a semi-quantitative method. To avoid this variability, we designed a software algorithm for counting B-lines. We compared the number of B-lines obtained in real-time by observers with three different levels of experience and by the software algorithm, and analyzed intra-rater variability in terms of the estimated number of B-lines in two successive examinations. Forty mechanically ventilated adult (≥ 18 years) intensive care unit patients were included in this prospective study. All patients underwent two consecutive ultrasound examinations for B-lines detection by three human observers (OB1 = high, OB2 = medium, OB3 = low level of experience) and by the software (OBS). Ultrasound scans were obtained on the anterior right and left thoracic side along the midclavicular line, in the second and fourth intercostal space; B-lines counting for each position lasted 10 s. To assess intra-observer variability, a second ultrasound scan was obtained 15-30 min after the first scan. For all lung zones, the intraclass correlation for B-lines counting between OB1 and OB2 was 0.663; between OB1 and OB3, 0.559; and between OB1 and OBS, 0.710. OBS had a better concordance coefficient (0.752) between the first and the second measurements than did the human observers. Our results show that the software algorithm for B-lines counting is a potentially promising alternative when observers have little lung ultrasound experience.