Individualized model of torso surface for the inverse problem of electrocardiology

Effects of torso mesh density and electrode distribution on the accuracy of electrocardiographic imaging during atrial fibrillation

Introduction: Electrocardiographic Imaging (ECGI) allows computing the electrical activity in the heart non-invasively using geometrical information of the patient and multiple body surface signals. In the present study we investigate the influence of the number of nodes of geometrical meshes and recording ECG electrodes distribution to compute ECGI during atrial fibrillation (AF).

Methods: Torso meshes from 100 to 2000 nodes heterogeneously and homogeneously distributed were compared. Signals from nine AF realistic mathematical simulations were used for computing the ECGI. Results for each torso mesh were compared with the ECGI computed with a 4,000 nodes reference torso. In addition, real AF recordings from 25 AF patients were used to compute ECGI in torso meshes from 100 to 1,000 nodes. Results were compared with a reference torso of 2000 nodes. Torsos were remeshed either by reducing the number of nodes while maximizing the overall shape preservation and then assigning the location of the electrodes as the closest node in the new mesh or by forcing the remesher to place a node at each electrode location. Correlation coefficients, relative difference measurements and relative difference of dominant frequencies were computed to evaluate the impact on signal morphology of each torso mesh.

Results: For remeshed torsos where electrodes match with a geometrical node in the mesh, all mesh densities presented similar results. On the other hand, in torsos with electrodes assigned to closest nodes in remeshed geometries performance metrics were dependent on mesh densities, with correlation coefficients ranging from 0.53 ± 0.06 to 0.92 ± 0.04 in simulations or from 0.42 ± 0.38 to 0.89 ± 0.2 in patients. Dominant frequency relative errors showed the same trend with values from 1.14 ± 0.26 to 0.55 ± 0.21 Hz in simulations and from 0.91 ± 0.56 to 0.45 ± 0.41 Hz in patients.

Conclusion: The effect of mesh density in ECGI is minimal when the location of the electrode is preserved as a node in the mesh. Torso meshes constructed without imposing electrodes to constitute nodes in the torso geometry should contain at least 400 nodes homogeneously distributed so that a distance between nodes is below 4 cm.

Transfer Learning From Simulations on a Reference Anatomy for ECGI in Personalized Cardiac Resynchronization Therapy

GOAL

Noninvasive cardiac electrophysiology (EP) model personalisation has raised interest for instance in the scope of predicting EP cardiac resynchronization therapy (CRT) response. However, the restricted clinical applicability of current methods is due in particular to the limitation to simple situations and the important computational cost.

METHODS

We propose in this manuscript an approach to tackle these two issues. First, we analyze more complex propagation patterns (multiple onsets and scar tissue) using relevance vector regression and shape dimensionality reduction on a large simulated database. Second, this learning is performed offline on a reference anatomy and transferred onto patient-specific anatomies in order to achieve fast personalized predictions online.

RESULTS

We evaluated our method on a dataset composed of 20 dyssynchrony patients with a total of 120 different cardiac cycles. The comparison with a commercially available electrocardiographic imaging (ECGI) method shows a good identification of the cardiac activation pattern. From the cardiac parameters estimated in sinus rhythm, we predicted five different paced patterns for each patient. The comparison with the body surface potential mappings (BSPM) measured during pacing and the ECGI method indicates a good predictive power.

CONCLUSION

We showed that learning offline from a large simulated database on a reference anatomy was able to capture the main cardiac EP characteristics from noninvasive measurements for fast patient-specific predictions.

SIGNIFICANCE

The fast CRT pacing predictions are a step forward to a noninvasive CRT patient selection and therapy optimisation, to help clinicians in these difficult tasks.

Detailed Anatomical and Electrophysiological Models of Human Atria and Torso for the Simulation of Atrial Activation

Atrial arrhythmias, and specifically atrial fibrillation (AF), induce rapid and irregular activation patterns that appear on the torso surface as abnormal P-waves in electrocardiograms and body surface potential maps (BSPM). In recent years both P-waves and the BSPM have been used to identify the mechanisms underlying AF, such as localizing ectopic foci or high-frequency rotors. However, the relationship between the activation of the different areas of the atria and the characteristics of the BSPM and P-wave signals are still far from being completely understood. In this work we developed a multi-scale framework, which combines a highly-detailed 3D atrial model and a torso model to study the relationship between atrial activation and surface signals in sinus rhythm. Using this multi scale model, it was revealed that the best places for recording P-waves are the frontal upper right and the frontal and rear left quadrants of the torso. Our results also suggest that only nine regions (of the twenty-one structures in which the atrial surface was divided) make a significant contribution to the BSPM and determine the main P-wave characteristics.

Noninvasive identification of two lesions with local repolarization changes using two dipoles in inverse solution simulation study

BACKGROUND

The method for inverse localization and identification of two distinct simultaneous lesions with changed repolarization in the ventricular myocardium (two-vessel disease) is proposed and its robustness to errors in input data is tested in this simulation study.

METHOD

The inverse solution was obtained from the difference between STT integral body surface potential map computed with repolarization changes and the STT integral map from normal activation. In a numerical model of ventricles 48 cases of two simultaneous lesions and 48 cases of a single lesion were modeled. The effect of the lesions was taken to be represented by two dipoles. The input data were disturbed by three types of added noise. Twenty three characteristics of every obtained inverse solution were defined and four of them were used as the features in discriminant analysis task distinguishing the correct inverse solutions identifying two lesions.

RESULTS

The mean localization error for identified two lesions was 1.1±0.7cm. The sensitivity and specificity of quadratic discriminant analysis with cross-validation and feature selection was higher than 90%.

CONCLUSIONS

The combination of the inverse solution with two dipoles and discriminant analysis allows the identification of two simultaneous lesions without a priori information about the number of lesions.