In-silico pace-mapping using a detailed whole torso model and implanted electronic device electrograms for more efficient ablation planning

Automated Localization of Focal Ventricular Tachycardia From Simulated Implanted Device Electrograms: A Combined Physics–AI Approach

“…The workflow of this study is summarized in Figure 1 . Briefly, this involved using a previously generated 3D torso model ( Monaci et al, 2020 ) (step 1) to rapidly simulate focal pacing across the LV within a RE environment (step 2). These simulated paced beats were combined with LF matrices computed on the standard 12-lead ECG electrodes and manufacturer-guided CIED right (RV) and left (LV) ventricular leads (step 3) to reproduce accurate 12-lead ECG and EGM recordings of the paced beats (step 4).…”

“…As in Monaci et al (2020) , a 3D torso model was generated from a computed tomography (CT) trans-catheter aortic valve implantation (TAVI) panning scan. The torso model included all major organs, with conductivities reported in Table 1 , and a detailed four-chamber heart, extracted from a separate cardiac CT scan.…”

“…To replicate focal ectopic VTs across the LV segments, ∼3767 randomly chosen paced beats—single stimuli, with a basic cycle length (BCL) of 400 ms—were simulated using a computationally efficient RE formulation ( Neic et al, 2017 ) within the Cardiac Arrhythmia Research Package (CARP) ( Vigmond et al, 2003 ), utilizing the 10 Tusscher ventricular cell model ( ten Tusscher et al, 2004 ). Intra- and extracellular tissue conductivities were tuned to achieve physiological QRSs ( Costa et al, 2013 ), comparable to equivalent pseudo bidomain simulations on a higher-resolution mesh ( Monaci et al, 2020 ). Intra- and extracellular conductivities were 0.1845S/m and 0.6628S/m along the fiber direction, respectively, and 0.0493S/m and 0.1769S/m transverse to it.…”

“…Their combination with the RE solutions produced high-fidelity 12-lead ECGs and EGM traces ( Figures 4E,F ) in ∼20 s (256 cores) for each paced location. Eight EGM vectors were chosen as the main EGM signals ( Monaci et al, 2020 ), and included far-field CAN-SVC, CAN-RV, and SVC-RV, and near-field RV tip-RV ring and each LV tip-RV tip. However, importantly, additional vector combinations (four for ECGs and eight for EGMs) were added to the standard signals to facilitate integration into the CNN algorithms (see section “CNNs Training and Testing”).…”

“…Recent clinical studies have demonstrated that stored EGM recordings of re-entrant VT episodes from implanted devices can be successfully used to guide the construction of pace-maps during an ablation procedure, with similar accuracy to the use of ECGs, but with the advantage of ensuring that the clinical VT is targeted ( Yoshida et al, 2010 ; Yokokawa et al, 2019 ). In our own recent work, we demonstrated how such EGM recordings might be utilized to perform patient-specific in silico pace-mapping ( Monaci et al, 2020 ), improving pre-procedural ablation planning for complex scar-related VTs. However, literature lacks further investigation on the power of EGM recordings for the localization of focal VTs, for which computational models can address and answer a variety of different questions, and their use in AI-based algorithms.…”