Personalized computational heart models with T1-mapped fibrotic remodeling predict sudden death risk in patients with hypertrophic cardiomyopathy
Curation statements for this article:-
Curated by eLife
Evaluation Summary:
This study fuses images from cardiac magnetic resonance imaging and T1-mapping to reconstruct 3D anatomical models of the heart from hypertrophic cardiomyopathy patients. Using the model, they investigated potential contributions of diffuse fibrosis to arrhythmogenesis of the heart model in response to focal stimulation. While not perfect, the computer model significantly outperforms other risk predictors, and highlights diffuse fibrosis as a possible underlying cause. This study will be of interest to clinicians and basic scientists involved in heart rhythm research.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
- Medicine (eLife)
Abstract
Hypertrophic cardiomyopathy (HCM) is associated with risk of sudden cardiac death (SCD) due to ventricular arrhythmias (VAs) arising from the proliferation of fibrosis in the heart. Current clinical risk stratification criteria inadequately identify at-risk patients in need of primary prevention of VA. Here, we use mechanistic computational modeling of the heart to analyze how HCM-specific remodeling promotes arrhythmogenesis and to develop a personalized strategy to forecast risk of VAs in these patients. We combine contrast-enhanced cardiac magnetic resonance imaging and T1 mapping data to construct digital replicas of HCM patient hearts that represent the patient-specific distribution of focal and diffuse fibrosis and evaluate the substrate propensity to VA. Our analysis indicates that the presence of diffuse fibrosis, which is rarely assessed in these patients, increases arrhythmogenic propensity. In forecasting future VA events in HCM patients, the imaging-based computational heart approach achieved 84.6%, 76.9%, and 80.1% sensitivity, specificity, and accuracy, respectively, and significantly outperformed current clinical risk predictors. This novel VA risk assessment may have the potential to prevent SCD and help deploy primary prevention appropriately in HCM patients.
Article activity feed
-
-
Author Response:
Reviewer #1 (Public Review):
This study fused images from CMR and T1 mapping to reconstruct 3D anatomical models of the heart for HCM patients. Using the model, they investigated potential contributions of diffusive fibrosis to arrhythmogenesis of the heart model in response to focal stimulus. They found that the diffusive fibrosis contributed to increased incidence of ventricular arrhythmias.
The study is of some interest. However, there are some concerns regarding its publication in its present form.
- Details are unclear about how the imaging segmentation and alignment were conducted. Especially when CMR and T1-mapping data were fused together, how the slice images were aligned as mismatch is of a challenge and can affect the simulation results and conclusion.
The short-axis LGE-CMR and post-contrast T1 …
Author Response:
Reviewer #1 (Public Review):
This study fused images from CMR and T1 mapping to reconstruct 3D anatomical models of the heart for HCM patients. Using the model, they investigated potential contributions of diffusive fibrosis to arrhythmogenesis of the heart model in response to focal stimulus. They found that the diffusive fibrosis contributed to increased incidence of ventricular arrhythmias.
The study is of some interest. However, there are some concerns regarding its publication in its present form.
- Details are unclear about how the imaging segmentation and alignment were conducted. Especially when CMR and T1-mapping data were fused together, how the slice images were aligned as mismatch is of a challenge and can affect the simulation results and conclusion.
The short-axis LGE-CMR and post-contrast T1 acquisitions for each patient were performed consecutively during a single scan as described by Chu et al., 2017. In our study, we used the z-axis coordinate of the post-contrast T1 map (a short-axis, single slice) to select the corresponding short-axis LGE-MRI slice with the same position and orientation of the ventricle. Each set of a post-contrast T1 map and a corresponding short-axis LGE-MRI slice was visually inspected for differences in anatomy, cardiac phase, and distribution of enhancement, and only images found to be in agreement by the radiologists were used in this study. Information along these lines is now provided in the manuscript (Geometrical Reconstruction).
The short-axis LGE-CMR stack was used to reconstruct the LV geometry (3D volume). The LV myocardium was segmented in the CardioViz3D software using a validated semi-automatic landmark-based method used in previous studies by our team. This has been clarified with additional detail and citations (Arevalo et al., 2016; Shade et al., 2020b; Cartoski et al., 2019 in the revised submission (Geometrical Reconstruction).
The LV myocardium of the single post-contrast T1 map was segmented using the same method as above. The signal intensity profile of the myocardium from the corresponding LGE-CMR slice was normalized to the intensity profile (relaxation times) of the T1 map myocardium. Using thresholds of 350 and 450 ms, as described in the manuscript, we calculated the resulting standard deviations from the mean of the low signal intensity region of the LGE-CMR slice. These new, personalized standard deviation thresholds were then applied to each LGE-CMR slice of the short-axis stack to produce the regions of focal scar and diffuse fibrosis in the virtual heart model. Clarification has been added (Geometrical Reconstruction) in the revised submission.
- It is unclear what is the spatial resolution of the CMR, and how the spatial resolution of about 330 micrometre was achieved for the finite element model.
The LGE-CMR resolution is 2x2x8 mm and the post-contrast T1 map resolution is 1.5x1.5x8 mm. In the revised manuscript, we have added the spatial resolution of the CMR images (Imaging Data).
As to the reviewer’s second question, the equations of action potential propagation in the heart (a partial differential equation describing current flow in electrically interconnected cells, coupled to a set of ordinary differential and algebraic equations describing transmembrane currents) are solved on a left ventricular finite element mesh constructed from the segmented images. The finite element tetrahedral mesh needs to be of resolution 300-400 um (average resolution of 355um in our study) to achieve a stable converging solution of the equations; numerous studies have established and validated this spatial resolution value.
To construct the finite element tetrahedral mesh from the segmented images at the needed spatial resolution, we used the Mimics Innovation Suite from Materialise. The software uses an input target finite element edge length and generates a computational mesh with a tight edge length distribution around the input value. We have provided information regarding how the spatial resolution of the finite element mesh was achieved (including references supporting the requirement for mesh resolution); (Geometrical Reconstruction).
- It is unclear how the incorporation of fibre structures was done and validated. Given that fact that at different stages of HCM and individual differences, the fibre structures are different in different subjects. Without consideration of this, conclusions based on the diffusive fibrosis are non-conclusive.
We do not assume that the fiber orientations are the same in each patient’s heart. Only the very general rules that fiber tracts follow in the left ventricle are the same among subjects, but the fiber orientations in each left ventricle are specific to the geometry of that ventricle.
Fiber orientations were assigned in each model on the basis of the individual geometry of the ventricles in the following manner: Fiber orientations were assigned to each individual ventricular computational mesh on a per-element basis using an efficient rule-based approach that we developed and extensively validated (see reference Bayer et al., 2012 in this manuscript); the approach is now a staple in our field and is used widely in patient-specific ventricular simulation studies. The fiber orientation methodology uses the Laplace–Dirichlet method to define transmural and apicobasal directions at every point in the patient-specific ventricular mesh. It then employs bi-directional spherical linear interpolation to assign fiber orientations based on a general set of fiber orientation properties (rules) derived from a large amount of histological and diffusion tensor MRI data. We have provided more detail in the revised manuscript (Geometrical Reconstruction).
- It is also unclear how the physiological model for the HCM was developed and validated for the patient-specific model.
The HCM-specific cell model used to represent regions of diffuse fibrosis in this study is a modification of the ten Tusscher human ventricular model (see citation in manuscript). Modifications were made to the ion channel kinetics in this model based on experimental data from human HCM cardiomyocytes reported by Coppini et al. 2013 (cited in the manuscript). In that study, measurements were made via whole-cell voltage and current clamp in cardiomyocytes collected during myectomy from patients with HCM from regions shown to contain substantial amounts of diffuse fibrosis. Specific changes included 107% increase of INaL maximal conductance, 19% increase of ICaL maximal conductance, 34% decrease of IKr maximal conductance, 27% decrease of IKs maximal conductance, 85% decrease of Ito maximal conductance, 15% decrease of IK1 maximal conductance, 34% increase of sodium-calcium exchanger (NCX) activity, and 43% reduction of Sarcoplasmic/Endoplasmic Reticulum Calcium ATPase (SERCA) activity. The net results of the changes to the cell model include increased action potential duration at 90% repolarization from 280 to 330 ms (+18%) and diminution of the notch after depolarization. We have provided more detail and citations in the revised manuscript (Electrophysiological Properties).
Reviewer #2 (Public Review):
The overall aims of this work are to use computer models of electrical activation to (i) understand how remodelling of structure and function in hypertrophic cardiomyopathy promotes ventricular arrhythmias, and (ii) to assess whether a model-based approach could be used to predict the risk of arrhythmias in specific patients.
The approach taken by the authors builds on previous work by this group, where a personalized mesh representing the ventricles is constructed from automated analysis of cardiac MRI. Models of human electrophysiology are then solved on this mesh with simulated pacing, to identify vulnerability to arrhythmias.
The major strength of this approach is that it presents an environment within which an investigation that may be technically difficult, time-consuming, or unethical in a patient can be undertaken to guide treatment or assess risk. It is very promising.
However, although the methodology used is sound, there are important assumptions that underpin this approach and limit the extent to which the outcomes are trustworthy. These include:
- MRI physics. The MR signal is produced from a finite volume of tissue, which is about an order of magnitude larger than the size of finite elements used in the computer model. Thus, the personalised mesh may not capture small scale features that could be important for initiation of arrhythmias.
The reviewer is correct – the MRI scan and thus the computational mesh constructed from the segmented images does not capture small scale features. It indeed is possible that small scale features could lead to arrhythmogenesis in these patients, but it is unknown and highly unlikely that they would dominate arrhythmogenesis in the HCM-remodeled substrate. Since we stratify each personalized HCM substrate as arrhythmogenic or not by inducibility of arrhythmia following pacing, we capture all the contributions of MRI-visualized structural remodeling to arrhythmogenesis. While there may be additional contributions from small scale heterogeneities to arrhythmogenesis, this would not change the patient’s positive stratification result. If small-scale heterogeneities not “seen” by the MRI are the main mechanism for arrhythmogenesis in these HCM patients, then there will be a large discrepancy between our results of simulations and the clinical outcome. As the reviewer can see from the results presented in the paper, this is not the case. Thus, while small heterogeneities might additionally contribute to arrhythmogenesis, they do not alter the predominant mechanism on which stratification is based. We have added new text in the Study Limitations.
- Cardiac mechanics. The mesh used to solve the computer model is static, whereas the heart contracts with every beat. Mechanical contraction not only changes the shape of the heart and the thickness of the ventricular wall, but also feeds back into electrical activity.
Again, the reviewer is correct in that assessment. However, it is important to emphasize that in our translational effort to bring computational modeling to the point-of-care and clinical decision making, there are simplifications that need to be made to personalized models to make the models computationally tractable while being clinically useful. No model will ever be correct in every respect – but we strive to develop clinically-useful models, that make better predictions than those made by the current clinical criteria. While there is certainly mechanoelectrical feedback, the arrhythmogenic propensity imparted by HCM remodeling on the ventricular substrate is the main factor that leads to a stratification of high risk. Again, if that was not the case, and mechanoelectrical feedback played a major role, there would be a large discrepancy between results of our simulations and the clinical outcome, which is clearly not the case in the paper.
Importantly, recently we published a paper in which we conducted a detailed simulation of ventricular arrhythmia in a patient-specific model with structural remodeling that incorporated full mechanics, hemodynamics and feedbacks (Salvador, M., Fedele, M., Africa, P. C., Sung, E., Dede, L., Prakosa, A., Chrispin, J., Trayanova, N. and Quarteroni, A. (2021) 'Electromechanical modeling of human ventricles with ischemic cardiomyopathy: numerical simulations in sinus rhythm and under arrhythmia', Computers in Biology and Medicine, 136, pp. 104674.). The model was very complex and took days to execute for one pacing site. Nonetheless, the simulations demonstrated that when a purely electrophysiological model was compared to this complex multi-physics model, in all cases where arrhythmia was inducible in the electrophysiological model, it was also inducible in the multi-physics model and vice versa. The only difference between the models was that in the case of the multi-physics model, the arrhythmia was unstable, while it was stable in the electrophysiological model. This indicates that including additional complexities in our patient-specific models (such as mechanoelectrical feedback) would not change the stratification of whether an arrhythmia will occur or not.
We have added text and citations in the Study Limitations along these lines.
- Population variability. The electrical model used in this study is a standard representation for the human ventricles. This is adjusted to capture some features of electrical activity in fibrotic regions, but these are not well characterised so assumptions are made. The patterns of electrical activation and recovery in the human heart vary from place to place within the human ventricles, with time within the same patient in response to external effects including autonomic activity, and from one patient to another. Hypertrophic cardiomyopathy is usually a progressive disease, so patterns of fibrosis may change over time.
The reviewer is correct -- we are making a number of assumptions in our HCM model. Until there is a way to characterize personalized electrophysiology non-invasively, the models must make these assumptions. Despite these assumptions, this model is able to make correct predictions and to stratify patients better than the current clinical criteria. Of course, there is always room for improvement, like in any model.
We do not state that assessment with our risk predictor is a one-time effort. Since HCM is a progressive disease, our risk predictor should be applied again when new patient imaging is acquired during follow up visits. These are already recommended for patients with HCM to monitor the changes in fibrotic remodeling. We already had text in the Discussion of the original submission stating this.
Nevertheless, this study has found evidence that diffuse fibrosis plays a role in the vulnerability to arrhythmias in hypertrophic cardiomyopathy, and found that, in this group of 26 patients, a model-based approach can provide a more accurate risk stratification than other methods based on patient clinical data.
Thank you very much for the comment.
-
Evaluation Summary:
This study fuses images from cardiac magnetic resonance imaging and T1-mapping to reconstruct 3D anatomical models of the heart from hypertrophic cardiomyopathy patients. Using the model, they investigated potential contributions of diffuse fibrosis to arrhythmogenesis of the heart model in response to focal stimulation. While not perfect, the computer model significantly outperforms other risk predictors, and highlights diffuse fibrosis as a possible underlying cause. This study will be of interest to clinicians and basic scientists involved in heart rhythm research.
(This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.)
-
Reviewer #1 (Public Review):
This study fused images from CMR and T1 mapping to reconstruct 3D anatomical models of the heart for HCM patients. Using the model, they investigated potential contributions of diffusive fibrosis to arrhythmogenesis of the heart model in response to focal stimulus. They found that the diffusive fibrosis contributed to increased incidence of ventricular arrhythmias.
The study is of some interest. However, there are some concerns regarding its publication in its present form.
Details are unclear about how the imaging segmentation and alignment were conducted. Especially when CMR and T1-mapping data were fused together, how the slice images were aligned as mismatch is of a challenge and can affect the simulation results and conclusion.
It is unclear what is the spatial resolution of the CMR, and how the spatial …
Reviewer #1 (Public Review):
This study fused images from CMR and T1 mapping to reconstruct 3D anatomical models of the heart for HCM patients. Using the model, they investigated potential contributions of diffusive fibrosis to arrhythmogenesis of the heart model in response to focal stimulus. They found that the diffusive fibrosis contributed to increased incidence of ventricular arrhythmias.
The study is of some interest. However, there are some concerns regarding its publication in its present form.
Details are unclear about how the imaging segmentation and alignment were conducted. Especially when CMR and T1-mapping data were fused together, how the slice images were aligned as mismatch is of a challenge and can affect the simulation results and conclusion.
It is unclear what is the spatial resolution of the CMR, and how the spatial resolution of about 330 micrometre was achieved for the finite element model.
It is unclear how the incorporation of fibre structures was done and validated. Given that fact that at different stages of HCM and individual differences, the fibre structures are different in different subjects. Without consideration of this, conclusions based on the diffusive fibrosis are non-conclusive.
It is also unclear how the physiological model for the HCM was developed and validated for the patient-specific model.
-
Reviewer #2 (Public Review):
The overall aims of this work are to use computer models of electrical activation to (i) understand how remodelling of structure and function in hypertrophic cardiomyopathy promotes ventricular arrhythmias, and (ii) to assess whether a model-based approach could be used to predict the risk of arrhythmias in specific patients.
The approach taken by the authors builds on previous work by this group, where a personalized mesh representing the ventricles is constructed from automated analysis of cardiac MRI. Models of human electrophysiology are then solved on this mesh with simulated pacing, to identify vulnerability to arrhythmias.
The major strength of this approach is that it presents an environment within which an investigation that may be technically difficult, time-consuming, or unethical in a patient can …
Reviewer #2 (Public Review):
The overall aims of this work are to use computer models of electrical activation to (i) understand how remodelling of structure and function in hypertrophic cardiomyopathy promotes ventricular arrhythmias, and (ii) to assess whether a model-based approach could be used to predict the risk of arrhythmias in specific patients.
The approach taken by the authors builds on previous work by this group, where a personalized mesh representing the ventricles is constructed from automated analysis of cardiac MRI. Models of human electrophysiology are then solved on this mesh with simulated pacing, to identify vulnerability to arrhythmias.
The major strength of this approach is that it presents an environment within which an investigation that may be technically difficult, time-consuming, or unethical in a patient can be undertaken to guide treatment or assess risk. It is very promising.
However, although the methodology used is sound, there are important assumptions that underpin this approach and limit the extent to which the outcomes are trustworthy. These include:
1. MRI physics. The MR signal is produced from a finite volume of tissue, which is about an order of magnitude larger than the size of finite elements used in the computer model. Thus, the personalised mesh may not capture small scale features that could be important for initiation of arrhythmias.
2. Cardiac mechanics. The mesh used to solve the computer model is static, whereas the heart contracts with every beat. Mechanical contraction not only changes the shape of the heart and the thickness of the ventricular wall, but also feeds back into electrical activity.
3. Population variability. The electrical model used in this study is a standard representation for the human ventricles. This is adjusted to capture some features of electrical activity in fibrotic regions, but these are not well characterised so assumptions are made. The patterns of electrical activation and recovery in the human heart vary from place to place within the human ventricles, with time within the same patient in response to external effects including autonomic activity, and from one patient to another. Hypertrophic cardiomyopathy is usually a progressive disease, so patterns of fibrosis may change over time.
Nevertheless, this study has found evidence that diffuse fibrosis plays a role in the vulnerability to arrhythmias in hypertrophic cardiomyopathy, and found that, in this group of 26 patients, a model-based approach can provide a more accurate risk stratification than other methods based on patient clinical data.
-
Reviewer #3 (Public Review):
In their paper the authors set out to develop a novel ventricular arrhythmia risk assessment in the setting of hypertrophic cardiomyopathy.
The authors combine contrast enhanced MRI with T1 mapping data to construct computer models of HCM patient hearts that include patient specific distribution of fibrosis. They then use these personalized hearts to assess the propensity for ventricular arrhythmias.
The others demonstrate using a computational approach that diffuse fibrosis increases vulnerability to ventricular arrhythmia. It's an important finding especially because diffuse fibrosis is not a parameter that is typically tracked in HCM patients.
The potential impact of the work described in the study is high. By accounting for diffuse fibrosis as a risk factor for ventricular arrhythmias in HCM patients, …
Reviewer #3 (Public Review):
In their paper the authors set out to develop a novel ventricular arrhythmia risk assessment in the setting of hypertrophic cardiomyopathy.
The authors combine contrast enhanced MRI with T1 mapping data to construct computer models of HCM patient hearts that include patient specific distribution of fibrosis. They then use these personalized hearts to assess the propensity for ventricular arrhythmias.
The others demonstrate using a computational approach that diffuse fibrosis increases vulnerability to ventricular arrhythmia. It's an important finding especially because diffuse fibrosis is not a parameter that is typically tracked in HCM patients.
The potential impact of the work described in the study is high. By accounting for diffuse fibrosis as a risk factor for ventricular arrhythmias in HCM patients, the authors demonstrate improved sensitivity, specificity and accuracy compared to other risk predictive parameters. It is feasible that as a result of the study that diffuse fibrosis may be tracked in these patients as an indicator of propensity to deadly arrhythmias.
-
