InVision Precision LVEF (LVEF)

InVision Precision LVEF is a software medical device that employs a machine learning algorithm with spatiotemporal pooling for semi-automated segmentation of the left ventricle from transthoracic echocardiograms to calculate Left Ventricular Ejection Fraction. The system aims to assist clinicians by providing adjustable, real-time LVEF estimations from DICOM images integrated within PACS, with its performance in estimating LVEF and segmentation accuracy validated by RMSD and Dice scores against cardiologist consensus.

The InVision Precision LVEF is a Software as a Medical Device (SaMD) designed to assist clinicians in cardiac evaluations by providing automated estimations of Left Ventricular Ejection Fraction (LVEF). It processes previously acquired transthoracic cardiac ultrasound images in DICOM format, allowing for image storage, manipulation, and measurement on various platforms including ultrasound devices, personal computers, or compatible DICOM-compliant PACS systems. Indicated for use in patients 22 years and older, the software is operated by sonographers and physicians to aid in the diagnostic review and analysis of echocardiographic data, particularly for LVEF assessment, by selecting appropriate echocardiogram videos and generating segmentations. The core of InVision Precision LVEF relies on a machine learning algorithm utilizing a spatiotemporal pooling network architecture to perform semi-automatic segmentation of the left ventricle from selected echocardiogram video views (single or biplane). This segmentation forms the basis for the LVEF calculation, which is presented as a numerical estimate with adjustable annotations that clinicians can modify, with the LVEF calculation updating in real-time. Performance was validated in a retrospective, multicenter study using data from various ultrasound equipment manufacturers, demonstrating an average Root Mean Square Deviation (RMSD) for biplane view LVEF of approximately 6.06 against ground truth, and Dice scores for A4C and A2C segmentations of 0.89 and 0.90 respectively, across diverse patient subgroups.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the 'spatiotemporal pooling' network architecture used by InVision Precision LVEF compare in performance and robustness to the 'simple pooling' used by the predicate device, particularly for challenging image qualities or rare cardiac conditions?; What specific training data characteristics (e.g., diversity of patient populations, range of LVEF values, image quality variations) were prioritized to ensure the generalizability and minimize bias in the machine learning model's LVEF estimations?; Given the semi-automated nature, what is the typical time saved or efficiency gained by clinicians using InVision Precision LVEF compared to manual LVEF calculation, and how frequently do users need to adjust the initial automated segmentations?; What are the performance metrics (RMSD, Dice score) for patient subgroups with particularly high or low LVEF values, or those with significant comorbidities that might affect cardiac image quality or morphology?; How does the system handle cases where the automated view selection algorithm fails to identify appropriate cardiac views, and what is the fallback mechanism for the user?