CI-452767-3 MACHINE LEARNING-POWERED, DEVICE-EMBEDDED HEART SOUND MEASUREMENT CAN OPTIMIZE AV DELAY IN CRT PATIENTS

Abstract

Listen

Translate article

Continuous ambulatory optimization of atrioventricular (AV) delay for cardiac resynchronization therapy (CRT) is mainly performed by electrical means. Development of an estimation model of cardiac function that uses a piezoelectric transducer embedded in an internal pulse generator to guide CRT optimization in patients. Electrocardiogram, left ventricular pressure (LVP) and heart sounds were simultaneously collected during the implantation procedures. A piezoelectric alarm transducer embedded in a modified CRT device facilitated recording heart sound in 22 patients undergoing a pacing protocol with AV delays varying from 60 to 330 ms. Post-hoc, machine learning was employed to produce a decision tree ensemble model capable of estimating maximal LVP (LVPmax) and maximal rate of rise of LVP (LVdP/dtmax) using heart-sound based features. To gauge the applicability of machine learning in AV delay optimization, polynomial curves were fitted to invasively measured and estimated values. The figure illustrates the concept of the study and an example of the invasively measured and heart sound-estimated AV delay optimization curves. In the dataset of ∼30,000 heartbeats, machine learning indicated S1 amplitude, S2 amplitude and S1 integral (or energy) as the most prominent features for AV delay optimization in terms of LVPmax and LVdP/dtmax. Machine learning facilitated categorical beat by beat estimation for absolute values of LVPmax (10 mmHg bin size) and LVdP/dtmax (100 mmHg/s bin size). The right panel illustrates the similar absolute values for LVPmax using 30 second averages. Estimated optimal AV delays using ∼30 second recordings were not significantly different from those measured using invasive LVP (difference: -5.6 ± 17.1 ms for LVPmax and +5.1 ± 46.7 ms for LVdP/dtmax). The difference in function between the estimated and measured optimal AV delays was small and not statically significant (1 ± 3mmHg for LVPmax and 9 ± 57mmHg/s for LVdP/dtmax). Heart sound sensors embedded in a CRT device, powered by a machine learning algorithm provide a reliable assessment of both optimal AV delays and absolute LVPmax and LVdP/dtmax in a clinical environment.