10:30
Heart
10:30
15 mins
|
Joint estimation of parameters in a cardiac tissue model using confirmatory factor analysis
Miao Sun, Natasja M.S. de Groot, Richard C. Hendriks
Abstract: Mathematical modelling of electrical activity in cardiac tissue is useful to investigate the electrophysiological properties of cardiac cells and to help understanding the mechanisms underlying cardiac arrhythmias. However, as these models are rather detailed and include many parameters, it is challenging to use them for inverse modelling and to identify the parameters of interest. These could be parameters such as the conductivity, local activation time, and anisotropy ratio. Existing methods often assume part of the parameters to be known a priori and estimate the parameters separately. We have recently proposed to use an efficient method called confirmatory factor analysis (CFA) to jointly estimate a subset of these parameters from electrogram measurements [1]. However, the method in [1] was limited to the use of a single heart beat and could only estimate the conductivity and the anisotropy ratio. In this work, we further extend this method to jointly estimate the conductivity, anisotropy ratio, as well as the activation time per cell using multiple heart beats and multiple frequencies from atrial epicardial electrograms (EGMs).
The EGMs that we use are measured using a high resolution mapping approach presented in [2], measuring the transmembrane potentials of atrial tissue at multiple locations simultaneously. To utilize the spatial information of the data, we use the cross-power spectral density matrix (CPSDM) to model the multi-electrode EGMs and apply CFA to the CPSDM model to jointly estimate the parameters. The identifiability conditions that need to be satisfied in the CFA problem are used to find the relationship between the desired resolution and the required amount of data. With the reasonable assumptions that the conductivity and the anisotropy parameters are constant across different frequencies and heart beats, we estimate these parameters using multiple frequencies and multiple heart beats simultaneously to easier satisfy these identifiability conditions.
The results of the simulated data show that using multiple heart beats increases the estimation accuracy of the parameters. An evaluation using clinical data shows that using multiple heart beats for parameter estimation helps to reduce the reconstruction errors of the clinical EGMs, which further demonstrates the robustness of the proposed method.
[1] M. Sun, N. M. S. de Groot, R. C. Hendriks, Cardiac tissue conductivity estimation using confirmatory factor analysis, Computers in Biology and Medicine, vol. 135, 2021.
[2] A. Yaksh, L. J. van der Does, C. Kik, P. Knops, F. B. Oei, P. C. van de Woestijne, J. A. Bekkers, A. J. Bogers, M. A. Allessie, and N. M. de Groot, “A novel intra-operative, high resolution atrial mapping approach,” Journal of Interventional Cardiac Electrophysiology, vol. 44, no. 3, pp. 221–225, 2015.
|
10:45
15 mins
|
On the design of a novel portable ECG device for the recording of A 4-precordial electrode arrangement
Alejandra Zepeda Echavarria, Rutger van de Leur, Nynke de Vries, Rien van der Zee, Joris Jaspers, Thierry Wildbergh, Pieter Doevendans, René van Es
Abstract: With more than 300 million electrocardiograms (ECGs) obtained annually worldwide, the ECG is a fundamental tool in the everyday practice of clinical medicine.12-lead ECGs are commonly used by physicians to diagnose, monitor, record and understand the electrical activity of the heart. Technological advances have allowed to bring medical devices to home environments.
To investigate if an arrangement of 4- precordial electrodes could provide enough evidence to detect ECG changes, we developed a smartphone-sized ECG recording device, the miniECG, aiming to detect the full spectrum of ECG abnormalities.
The miniECG has the following components: four dry electrodes, a microcontroller unit, and an app. The four electrodes are stainless steel electrodes with peaks that could enable a low impedance contact. The design of the microcontroller unit is based on the ADS1298 (Texas Instruments, USA). Eight differential channels are recorded for 10 seconds, with a 24-bit resolution and a programmable gain up to 12V/V. The recordings and connection to app via Bluetooth is managed by the microcontroller NRF52840 (Nordic Semiconductor, Norway) featuring an Arm® Cortex ® -M4F CPU. The app designed to start and collect data on the recordings was designed on React Native. Finally, the high-level software for the processing and visualization of the ECG data was developed in python.
A first clinical study to determine if the miniECG is capable to detect ST-elevation has been ongoing from May 2021 at UMC Utrecht and Meander Medical Center. A total of 213 patients were included, 21 of whom demonstrated ST-elevation on their 12-lead ECG. In 18 of these patients the ST-elevation was also present on the miniECG recording, in the other 3, ST-elevation had also resolved on the follow up 12 lead ECG. Further analysis showed that the miniECG can record anterior, lateral, and inferior ST-elevation.
The miniECG can record a high-quality multi-lead ECG. Preliminary analysis of clinical study data shows that the miniECG can accurately detect ischemic ST-elevation in patients with inferior, lateral, and anterior myocardial infarction. Further research is required to demonstrate non-inferiority of the miniECG to the standard 12-lead ECG in the detection of other common cardiac (ab)normalities.
|
11:00
15 mins
|
Sex differences in drug-induced arrhythmogenesis
Mathias Peirlinck, Francisco Sahli Costabal, Ellen Kuhl
Abstract: Sex-specific electrocardiogram studies show that the electrical activity in the heart varies significantly between men and women. Recent evidence suggests that these differences also lead to differing responses to drugs. More specifically, women were found to be at least twice as likely to develop drug-induced arrhythmia with potentially fatal consequences. Yet, the sex-specific differences in drug-induced arrhythmogenesis remain poorly understood.
Here we integrate multiscale modeling and machine learning techniques to gain mechanistic insight into the sex-specific origin of drug-induced cardiac arrhythmia at differing drug concentrations. We set up male and female multiscale exposure-response simulators accounting for sex differences in subcellular ion channel activity, tissue-level conductivity, and organ-scale geometry. To quantify critical drug concentrations, we identify the most important ion channels that trigger arrhythmogenesis for each sex individually and subsequently train a sex-specific multi-fidelity arrhythmogenic risk classifier.
Our study reveals that sex differences in multiscale cardiac electrophysiology trigger longer QT-intervals in women than in men. Concomitantly, our risk classifiers predict significantly lower critical drug concentrations for women than for men. For dofetilide, a high-risk drug, we found the female critical drug concentration to be seven times lower than for men.
Our results emphasize the importance of including sex as an independent biological variable in risk assessment during drug development. Acknowledging and understanding sex differences in drug safety evaluation is critical when developing novel therapeutic treatments on a personalized basis. The general trends of this study have significant implications on the development of safe and efficacious new drugs and the prescription of existing drugs in combination with other drugs.
|
11:15
15 mins
|
Light weight deep neural network model for left ventricle segmentation in ultrasound imaging
Navchetan Awasthi, Lars Vermeer, Louis S. Fixsen, Josien P. W. Pluim, Richard G. P. Lopata
Abstract: Segmentation of ultrasound images of the left ventricle is an important task for assessment of geometry and cardiac deformation imaging. Manual segmentation or corrections by clinicians are typically needed as the conventional techniques do not give accurate results, often leading to over- and under-segmentation of the desired structure. Deep Learning (DL) based methods have shown to be promising techniques for segmentation, classification, detection of structures, and improving the reconstruction quality of images. Various architectures have been proposed for segmentation of medical images, including echocardiography, with extremely high accuracy and robustness. These models are heavy in terms of parameters, consume considerable time in inference, and can therefore not be deployed in point-of-care applications, e.g., on a mobile device.
First, we proposed a splitblock which utilizes multidilation depthwise separable convolutions and factorized convolutions. Utilizing this splitblock, we proposed a deep neural network – LVNet, which is a light-weight network. Here, a canine dataset is analysed, with ultrasound images acquired before and after left bundle block was introduced. The short axis view of these ultrasound images are acquired using a GE PA2-5 phased array transducer (3-MHz centre frequency, 75 degree opening angle with 90 frames per second). Manual annotations were performed by trained observers to generate the ground truth segmentations for both cases. The proposed model was compared to well-known deep learning models - UNet, MiniNetV2, and FCdDN.
The proposed model requires approximately 250 thousand parameters as compared to a UNet model which requires around 7 million parameters. Compared to the UNet model, an improvement in DICE score performance as high as 5% for segmentation with papillary muscles was found, while showcasing an improvement of 18.5 % when the papillary muscles are excluded. The model proposed also outperformed the MiniNetV2, and FCdDN models in terms of accuracy and model size. The result of this study warrants the use of this DL architecture for future use in point of care devices.
|
11:30
15 mins
|
Left atrial strain: a sensitive whistleblower lost in translation
Tim van Loon, Nick van Osta, Tijmen Koopsen, Joost Lumens
Abstract: Background: Left atrial (LA) function measured by peak/reservoir strain (PLAS) has been shown to be of prognostic value in various cardiac pathologies, including heart failure (HF). However, the underlying mechanisms leading to reduced PLAS remain poorly understood.
Purpose: To investigate how various combinations of left ventricular (LV) systolic-, diastolic, and LA failure affect PLAS.
Method: The CircAdapt lumped-parameter model of the human heart and circulation is used to simulate the abovementioned LV and LA substrates, starting from a healthy reference simulation. LV systolic failure was simulated by a 25% decrease of LV contractility and LV eccentric hypertrophy (simultaneous increase of LV wall mass and area by 30%). LV diastolic failure was simulated by increasing LV myocardial stiffness by 100%, impairing LV relaxation function (increasing relaxation time constant from 35ms to 60ms, and LV concentric hypertrophy (increasing LV wall mass by 30%). LA failure was simulated by a fully non-contractile LA myocardium.
Results: LV systolic failure was characterized by reduced LV GLS (21 to 12%) and LVEF (54 to 34%), with marginal increase in PLAS (42 to 46%). LV diastolic failure was characterized by reduced PLAS (42 to 23%) with LA dilation (i.e. peak LA volume: 50 to 80 ml). Furthermore, LV concentric hypertrophy increased LVEF (54 to 64%). LA failure led to LA dilation and reduction of PLAS, regardless of LV substrate. Lastly, LV diastolic dysfunction and LA failure did not affect LV GLS.
Conclusion: Simulations showed that acute LV systolic failure hardly affected PLAS, but predominantly decreased LV GLS and LVEF. On the contrary, both LV diastolic dysfunction and LA failure reduced PLAS, rather than LV GLS and LVEF. Hence, LA strain can be considered as a whistleblower sensitive to various LV and/or LA abnormalities, which may explain why it is a rather nonspecific diagnostic marker.
|
11:45
15 mins
|
Cardiac strain imaging using bistatic dual-probe ultrafast ultrasound
Peilu Liu, Hans-Martin Schwab, Richard Lopata
Abstract: Multi-perspective ultrafast ultrasound (US) has been introduced to improve cardiac strain imaging (Liu et al., IUS 2019), where two probes take turns to transmit and receive US signals separately (monostatic imaging). We now propose to leverage the full capability of this imaging system by additionally receiving and reconstructing US signals from the non-transmitting probe to obtain two extra trans-probe datasets (bistatic imaging). This study explores the use of multi-perspective ultrafast bistatic US imaging to further improve cardiac image contrast, resolution, motion tracking and strain estimation.
In an ex-vivo experiment of a beating porcine heart (LifeTec, NL), parasternal short axis views of the LV were obtained by two phased array probes (P4-2v, Verasonics) at a frame rate of 170 frames per second. An image intensity based fully automatic registration algorithm was designed to precisely register the dual-probe images and reconstruct the trans-probe images. To suppress trans-probe artifacts, US signals that directly transmitted from the second probe without reflection were removed in the channel data and a high pass filter was applied on radio frequency data. A fusion algorithm based on the discrete wavelet transform and regional energy fusion rule, was developed to optimally fuse the four image sets. First- and second-order speckle statistics were computed, and speckle tracking and axial displacements compounding were performed. The performances of image quality, motion tracking and strain estimation were compared for single probe (SP), dual-probe monostatic (DM) and dual-probe bistatic (DB) imaging, respectively.
Compared to SP images, less overlap of the histograms (-49%) between myocardium and endocardium, higher contrast-to-noise ratio (CNR, +67%), and improved and more isotropic resolution (eccentricity, -15%; lateral resolution, -0.26mm) were found in DB images. DB imaging reduced motion tracking error by 35%, improved elastographic signal-to-noise-ratio (SNRe) by 224% and increased strain magnitude by 267% (from 0.06 to 0.22) for radial strain. Compared to DM imaging, DB imaging improved the radial SNRe (+40.1%) and strain magnitude (+13%), in regions where strain was parallel to the axial direction of trans-probe images.
|
|