Ultra-high frequency lung ultrasound in preterm neonates: a test validation study on interpretation agreement and...

To verify if increasing frequency, through the use of ultra-high frequency transducers, has an impact on lung ultrasound pattern recognition.

Test validation study.

Tertiary academic referral neonatal intensive care unit.

Neonates admitted with respiratory distress signs.

Lung ultrasound performed with four micro-linear probes (10, 15, 20 and 22 MHz), in random order. Anonymised images (600 dpi) were randomly included in a pictorial database: physicians with different lung ultrasound experience (beginners (n=7), competents (n=6), experts (n=5)) blindly assessed it. Conformity and reliability of interpretation were analysed using intraclass correlation coefficient (ICC), area under the curve (AUC) of the multi-class ROC analysis, correlation and multivariate linear regressions (adjusting for frequency, expertise and their interaction).

A (0-3) score based on classical lung ultrasound semiology was given to each image as done in the clinical routine.

ICC (0.902 (95% CI: 0.862 to 0.936), p<0.001) and AUC (0.948, p<0.001) on the whole pictorial database (48 images acquired on 12 neonates), and irrespective of the frequency and physicians' expertise, were excellent. Physicians detected more B-lines with increasing frequency: there was a positive correlation between score and frequency (ρ=0.117, p=0.001); multivariate analysis confirmed the score to be higher using 22 MHz-probes (β=0.36 (0.02-0.7), p=0.041).

Overall conformity and reliability of interpretations of lung ultrasound patterns were excellent. There were differences in the identification of the B-patterns and severe B-patterns as increasing probe frequency is associated with higher score given to these patterns.