Machine learning based automated dynamic quantification of left heart chamber volumes

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Echocardiography

Background:

• Automated volumetric analysis of 3D echocardiographic (3DE) datasets provides accurate left ventricular (LV) and left atrial (LA) volume measurements.
• Machine learning (ML) has been applied to automatically generate LV and LA volume-time curves throughout the cardiac cycle.

Purpose of the Study:

• To validate ejection and filling parameters derived from ML-based 3DE volume-time curves.
• To compare ML-derived parameters against established reference techniques like conventional 3DE and cardiac magnetic resonance (CMR).

Main Methods:

• 20 patients underwent 3DE and cardiac magnetic resonance (CMR) imaging.
• LV and LA volume-time curves were generated using an ML algorithm (Philips HeartModel).
• Results were compared against conventional 3DE analysis (TomTec) and manual CMR tracing.

Main Results:

• The ML algorithm generated volume-time curves rapidly (35 ± 17 seconds) compared to conventional 3DE (3.6 ± 0.9 minutes) and CMR (96 ± 14 minutes).
• Minor manual correction of LV/LA borders was needed in a small percentage of cases (4/20 and 5/20, respectively).
• Ejection and filling parameters showed no significant inter-technique differences, with small biases observed via Bland-Altman analysis.

Conclusions:

• The automated ML algorithm accurately and quickly measures dynamic LV and LA volumes and analyzes ejection/filling parameters.
• This technology has the potential to increase the clinical utilization of 3DE imaging due to its efficiency.