Cardiac Natural Mechanical Wave Detection and Speed Estimation Using Deep Learning-Based 2-D Ultrasound Imaging: A Feasibility Study.
Authors
Affiliations (5)
Affiliations (5)
- Centre de Résonance Magnétique des Systèmes Biologiques, Bordeaux, France. Electronic address: [email protected].
- Sichuan University, School of Cyber Science and Engineering, Chengdu, China.
- CREATIS, Université Claude Bernard Lyon 1, Lyon, France.
- Polytechnique Montréal, Provost Lab, Montréal, Canada.
- Centre de Résonance Magnétique des Systèmes Biologiques, Bordeaux, France.
Abstract
Cardiac natural mechanical wave (NMW) propagation speed is a marker of the mechanical properties and pathological state of tissue, and its estimation relies on the ability to coherently sample small tissue displacements at a high imaging frame rate (above 800 fps). As the imaging frame rate increases, image quality is altered. Recently, a deep learning network designed specifically for compounding high-frame rate beam-formed data has proven to be efficient at image quality improvement. In this article, we aim to demonstrate that NMWs can be detected on deep learning compounded data and that their wave speed can be accurately estimated. Using a clutter filter wave imaging and spatiotemporal gradient on estimated wave time of flight, we demonstrated the ability to detect and track NMWs on deep learning compounded (DLC) data. We evaluated their wave speed in four healthy volunteers in different regions of interest, examined temporal signals and compared the velocity distribution across the entire myocardium. The atrial kick wave was detected in all four volunteers' DLC data, with atrial kick wave speed values estimated in the range 1-2m.s<sup>-1</sup>. Differences in wave speed distribution compared with coherent compounding are also highlighted. This work demonstrates the ability to detect and estimate NMW speed on deep learning-based ultrasound imaging. It also paves the way for neural network improvement to keep temporal information intact, as well as allow a more robust detection and wave speed estimation of NMWs on DLC data.