The purpose of this study is to validate the efficacy of Kosmos PLAX EF, an AI-assisted workflow that uses a cardiac ultrasound clip in Parasternal Long Axis (PLAX) cardiac view to calculate Left Ventricular Ejection Fraction (LVEF) using the Teichholz method. This study will compare the algorithm-generated measurements to measurements performed by hand by expert echocardiographers. The primary objectives of the study are to demonstrate that the algorithm measurements are statistically interchangeable with measurements from a reference standard derived from human expert readers. The primary hypotheses are that the algorithm measurements of left ventricular internal dimensions at 1) end-diastole (LVIDd) and 2) end-systole (LVIDs), respectively, will be statistically non-inferior to the reference standard. The secondary objective is to demonstrate that the algorithm-calculated LVEF is statistically interchangeable with the LVEF calculated by human readers. The secondary hypothesis is that the LVEF calculated from algorithm measurements will be statistically non-inferior to the reference standard. The study design involves the compilation of de-identified, retrospective PLAX-view clips acquired using the Kosmos diagnostic ultrasound system. This data was collected from two previous studies, and was not used in the training of the PLAX EF AI algorithm, or in the ECHO-300 PLAX EF pilot study.
Study Type
OBSERVATIONAL
Enrollment
416
All de-identified PLAX (Parasternal Long Axis) echocardiographic clips, along with corresponding demographic information, was collected from previously collected datasets. No additional imaging, interventions, or participant contact was performed as part of this study. Three independent, board-certified expert readers, blinded to automated outputs and to each other's assessments, manually measured LVIDd and LVIDs following ASE guidelines to produce the Reference Standard Measurements. The Kosmos EF algorithm processed each clip to generate automated LVIDd, LVIDs, and computed EF measurements to create the Automated Measurements. Automated measurements were compared to reference standard using an MAE-based methodology.
Memorial Healthcare System
Hollywood, Florida, United States
LVIDd Non-Inferiority Test
A panel of three expert readers will independently measure LVIDd on each study clip. The automated LVIDd measurement will be compared to the reference standard using MAE. Using the Mean Absolute Error (MAE) methodology, the automated LVIDd measurement error relative to expert readers will be non-inferior to human inter-reader variability, defined as the upper bound of the 95% bootstrap confidence interval for the median relative MAE difference being less than +25%.
Time frame: Following data acquisition (2 months)
LVIDs Non-Inferiority Test
A panel of three expert readers will independently measure LVIDs on each study clip. The automated LVIDs measurement will be compared to the reference standard using the MAE-based methodology. Using the Mean Absolute Error (MAE) methodology, the automated LVIDs measurement error relative to expert readers will be non-inferior to human inter-reader variability, defined as the upper bound of the 95% bootstrap confidence interval for the median relative MAE difference being less than +25%.
Time frame: Following data acquisition (2 months)
EF Non-Inferiority Test
EF will be computed from the automated LVIDd and LVIDs measurements using the Teicholz formula. The computed EF will be compared to the reference standard EF using the MAE-based methodology. Using the Mean Absolute Error (MAE) methodology, the computed EF error relative to expert readers will be non-inferior to human inter-reader variability, defined as the upper bound of the 95% bootstrap confidence interval for the median relative MAE difference being less than +25%.
Time frame: Following data acquisition (2 months)
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.