Aim. In recent years noninvasive electroanatomical mapping (EAM) techniques have been widely distributed. Using medical imaging data such as computer tomography (CT), magnetic resonance imaging (MRI) and ECG records provides powerful instruments for clinicians. Recently, we developed an algorithm for noninvasive EAM based on automatically segmented cardiac anatomy and 12-lead ECG data. However, this approach was developed for patients without extensive fibrosis in the heart. In this work, we have created an advanced algorithm for noninvasive EAM including fibrosis data.
Methods. In the scope of this study, the dataset of 15 patients with cardiac CT, late gadolinium enhancement (LGE) MRI, and 12-lead ECG were used. Next, based on CT images, 3D models of heart, lungs, and torso anatomies were automatically reconstructed using a previously published approach. A semi-automatic segmentation of MRI LGE data was performed for the reconstruction of fibrosis 3D geometry which was then aligned with 3D heart models by experts. For each case, the EAM was estimated using the early proposed method. Then, the modification of this method that includes a novel sampling method for fibrosis points and an advanced neural network-based mathematical model for ECG parameters. Finally, noninvasive maps with and without fibrosis in all the cases have been compared using Spearman correlation (R).
Results. The analyzed approaches showed the mean R 0.81±0.08 for the EAMs. The mean R between the modeled ECGs without fibrosis accounting and the recorded ECGs was 0.74±0.12. The mean R between the modelled ECGs with fibrosis accounting and recorded ECGs was 0.86±0.13. The mean R between the modelled ECGs with and without fibrosis was 0.89±0.07.
Conclusion. The results demonstrate that integrating fibrosis data into our algorithms significantly enhances correlation with originally recorded ECG data and improves the diagnostic accuracy of noninvasive EAM.