Epicardial Mapping from Venous Catheter Measurements,
Body Surface Potential Maps, and an Electrocardiographic Inverse Solution
Epicardial Mapping from Venous Catheter Measurements,
Body Surface Potential Maps, and an Electrocardiographic Inverse Solution
Robert S. MacLeod, Yesim Serinagaoglu*, Bulent Yilmaz,
and Dana H. Brooks*
NEH Cardiovascular Research and Training Institute (CVRTI), U of Utah*Communications and Digital Signal Processing (CDSP), Northeastern U
Combining Information SourcesCombining Information Sources
Forward
Inverse
Venous Catheter Based Mapping
Venous Catheter Based Mapping
Statistical EstimationStatistical Estimation
EstimationMatrix
Training Data
CovarianceMatrix
* =
Sparse Test DataEstimate
Estimated Activation MapsEstimated Activation MapsOriginal Mixed
RV-onlyLV-only
• Training set composition
• Lead selection
Augmented Inverse ProblemAugmented Inverse Problem
Forward
Inverse
Torso geometry
Body-Surface Potentials
Sparse Epicardial Potentials
Inverse Solution
Epicardial Map
+
+
+
Subtraction ApproachSubtraction ApproachKnownUnknown
Inverse(Tikhonov)
1) Subtract known epicardial potentials
2) Solve reduced inverse problem
Epicardial EstimationEpicardial Estimation
t1 t2 ….. ….tN
. . .
Estimation
Combined EstimationCombined Estimation
Estimation
Bayesian ApproachBayesian Approach
Inverse(MAP)
Hybrid ApproachHybrid Approach
Inverse(MAP)Estimate
Tank/Heart GeometryTank/Heart Geometry
Test Lead SetsTest Lead Sets
Anterior
Posterior
42 leads 21 leads 10 leads
Simulation StudySimulation Study
• 490 lead measured sock data
• Surrogate catheter potentials– 42 sites– + Gaussian noise
• Torso potentials– Calculated noise-free using forward model– + Gaussian noise
Leave-One-Experiment Out Protocol
Leave-One-Experiment Out Protocol
Training Data
Dec 13, 2000
Test Data
Dec 18, 2000
Dec 13, 2000
LV Paced Beats
Mixed Paced Beats
LV Pacing (LV-23 ms) LV Pacing (LV-23 ms) Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-MEpiP-23 ms
LV Pacing (LV-38 ms)LV Pacing (LV-38 ms)Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-MEpiP-38 ms
LV Pacing (LV-47 ms)LV Pacing (LV-47 ms)Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-MEpiP-47 ms
LV Pacing (Mixed-23 ms)LV Pacing (Mixed-23 ms)Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-RV-MEpiP-23 ms
LV Pacing (Mixed-38 ms)LV Pacing (Mixed-38 ms)Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-RV-MEpiP-38 ms
LV Pacing (Mixed-47 ms)LV Pacing (Mixed-47 ms)Orig Subt MAP
Est Epi Est All Hybrid MAP
31: LV-RV-MEpiP-47 ms
Relative Error (31-LV)Relative Error (31-LV)
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4R
elat
ive
Err
or
Time [ms]
MAPEstim-Epi
Estim-AllMAP-HybridSubtr
Relative Error (31-Mixed)Relative Error (31-Mixed)
0 10 20 30 40 50 60 70 80 90 1000
0.2
0.4
0.6
0.8
1
1.2
1.4MAPEstim-EpiEstim-AllMAP-HybridSubtr
Rel
ativ
e E
rro
r
Time [ms]
Estimation FindingsEstimation Findings
• Estimation alone: noisy, unstable results
• Estimation + inverse: smoothing improves stability
Inverse Solution FindingsInverse Solution Findings
• All solutions better than simple Tikhonov
• MAP usually improved with addition of catheter measurements (Hybrid MAP)
Role of Statistics (Training)Role of Statistics (Training)
• Generally helps
• But can add artifacts, e.g., spurious breakthroughs or wavefronts
• Torso potentials can reduce artifacts
ConclusionConclusion
More (bigger) is better…..
… but not always