Accounting for uncertainty in computational models of ventricular tachycardia improves ablation guidance

Figures

• 

  Fig. 1. Pipeline showing generation of the ablation target heat map. BiV models are created from segmented CCT images. Different thresholds of LVWT are used to define chronic scars to create multiple model ‘instances’ for each patient. Application of our rapid re-entrant circuit path-finding approach (VITA) uncovers multiple uVT circuits in each model instance, with simulation of corresponding ECG signatures. Correlation of simulated ECGs with the clinical target ECG VT template identifies highly correlated VT circuits. Corresponding isthmuses of these highly correlated circuits are spatially combined to generate the SATHe Map, intrinsically incorporating uncertainty in scar definition from clinical images.

• 

  Fig. 2. Example of scar thresholding in a representative patient. LVWT map derived from CT-based model (left). Three example model instances formed by selecting scar thresholds at LVWT < 3 mm, <4 mm and <5 mm producing markedly different scar morphologies (center). uVT circuits uncovered in each model instance (scar shown in gray) demonstrating the substantial variability in arrhythmogenic substrate between model instances (right).

• 

  Fig. 3. Distribution of induced uVTs and the maximum correlation with the clinical VT across model instances for all patients. a, Number of uVTs detected across scar-BZ model instances for 11 patients. Bars correspond to scar thresholds in 0.5-mm increments around the threshold yielding the peak CC. Three increments of scar threshold above the model instance yielding the peak CC and three below the peak are displayed. The central bar representing the model instance with the peak CC showing the value of the CC matching the clinical VT above it. The corresponding specific scar-BZ threshold levels for this model instance are shown in rows at the top (red, scar; yellow, BZ). Note: some model instances yielded zero uVTs, specifically: 2, 3, 1, 2 and 4 for patients 1–5, respectively. Patients 6–11 had no zero uVTs for any model instance shown. b, Similar plot to a, showing instead the explicit max CC values obtained within each model instance with the clinical VT target. Source data

• 

  Fig. 4. Simulated activation maps and ECG comparisons for three patients at maximum CC value. Simulated VT circuits and corresponding activation maps are shown alongside overlays of simulated (blue) and clinical (red) ECGs. Mean CC values reported correspond to averages over the top ten traces (as described in Methods, ‘Quantitative ECG comparison’ section).

• 

  Fig. 5. Scatter-plot of CCs for all simulated uVTs across all scar-BZ model instances in all patients. Each point denotes the CC for a uVT against the clinical ECG target. Maximum CCs across patients range from 0.75 to 0.91. As in Fig. 3, the scar-BZ threshold model instance yielding the max CC is annotated for each patient, and overlap counts (color bar) indicate the number of distinct uVTs sharing similar CCs. Source data

• 

  Fig. 6. Comparison of SATHe Maps with bipolar voltage map, ablation lesions and LAT maps of VT circuits for three patients. Bipolar voltage and VT LAT maps are plotted on CARTO shell geometries following registration with computational model using ultrasound-derived anatomical fiducials (Methods, ‘Comparison with clinical data’ section), shown in the shell registration panel. LVWT maps (far right) are plotted on the LV endocardial model surface. White dashed highlighted areas in bipolar voltage maps show approximate locations of hotspot areas in SATHe Maps following registration between geometries. Limits on the bipolar voltage maps highlight dense scar in red (<0.5 mV) and healthy myocardium in purple (>1.5 mV). Black arrows in VT LAT maps show approximate locations of diastolic isthmuses delineating the activation pathway (solid line) and hidden/subsurface activation pathway (dashed line) characteristic of transitions between early-to-late activation in close spatial proximity.

• 

  Fig. 7. Comparison of SATHe Maps with bipolar voltage map, ablation lesions and PaSo Maps for four patients in whom VT mapping was not possible. Bipolar voltage, lesions, registration and LVWT maps are the same as in Fig. 6. PaSo points are plotted following mapping back on to the computational model with colors indicating the pace-map correlation matching the target clinical ECG.

• 

  Fig. 8. Comparison of SATHe Maps with clinical data in the three patients with relatively little pathologically thinned LV wall. a, Patient 3 comparison with mapped VT circuit, similar to Fig. 5. b, Patients 5 and 8 comparison with PaSo Maps, similar to Fig. 7.

• 

  Extended Data Fig. 1. Computational workflow. Computational model generation and identification of relevant VT circuits. (a) Illustration of tetrahedral finite element mesh of BiV geometry. (b) Using LVWT for substrate characterization and generating model instances. (c) Identification of relevant VT circuit by correlating with clinically induced VT.