Category # of subjects
<25 kg/m
2
27
25-30 kg/m
2
22
30 kg/m
2
or higher 48
Ethnicity
5
Hispanic or Latino 2
NOT Hispanic or Latino 91
Unknown / Not Reported 4
Race
5
American Indian/Alaska Native 1
Black or African American 22
White 73
Unknown / Not Reported 1
B-line Count Performance: In both studies the inter-rater correlation coefficient (ICC) was calculated between the
Auto B-line Counter’s B-line count predictions and the ground truth. Ground truth was defined as the median of 9
expert annotators on the same set of cines. Both tests exceeded the performance goal of demonstrating an ICC
above a lower bound of 0.75. The performance target was derived from published literature
6
.
Acceptance Criteria ICC 95% CI
Study 1 Results
ICC ≥ 0.75
0.899 [0.867, 0.92]
Study 2 Results 0.85 [0.78, 0.90]
B-line Count Subgroup Analysis (Study 2)
Study 2 assessed the generalizability of the Auto B-line Counter across a wide range of clinically meaningful patient
subgroups (age, gender, BMI, ethnicity, and race). The tool performed similarly across all subgroups.
B-line Visualization (aka B-line Segmentation) Performance: Using Study 1 only, the degree of overlap in
localizing the position of B-lines was assessed using the Dice Similarity Coefficient (DSC) between the centroid-
paired segmentation from the Auto B-line Counter tool and the segmentation from the Ground Truth was calculated.
Ground truth for B-line segmentation was determined using 7 expert annotators. The DSC was calculated between
a B-line identified by the tool and a ground truth B-line that had complete or partial overlap, or abutted against one
another with no overlap. Study 1 exceeded the performance goal of demonstrating the DSC was equal or larger than
0.52. The performance target was derived from published literature
7
.
Acceptance Criteria
DSC 95% CI
Study 1 Results DSC ≥ 0.52 0.82 [0.78, 0.876]
10.2. Automatically Estimating Ejection Fractions
NOTE
The Simpson's Ejection Fraction tool is not available in the United States.
6
This approach follows that of an analysis of an AI/ML-based B-line counter algorithm described by Moore et al., “Interobserver Agreement
and Correlation of an Automated Algorithm for B-Line Identification and Quantification With Expert Sonologist Review in a Handheld Ultrasound
Device,” J Ultrasound Med 2021.
7
Derived from two papers: 1) Mason, Harry et al. “Lung Ultrasound Segmentation and Adaptation between COVID-19 and Community-Acquired
Pneumonia,” 2021, Accepted to MICCAI ASMUS Workshop (https://doi.org/10.48550/arXiv.2108.03138). 2) Roy, S. et al., "Deep Learning for
Classification and Localization of COVID-19 Markers in Point-of-Care Lung Ultrasound," in IEEE Transactions on Medical Imaging, vol. 39, no. 8,
pp. 2676-2687, Aug. 2020, doi: 10.1109/TMI.2020.2994459.
Automatically Estimating Ejection Fractions
AI-Assisted Tools 58
To Next Page
