m24 · Interpolation, activation maps and isochrones
Education only. Not for diagnosis, treatment or any clinical decision.
In 60 seconds
A mapping system knows activation times only where it recorded them; the colored activation map in between is interpolated. Linear interpolation on a triangulation is exact for linear fields, and its error grows as the square of the point spacing times the curvature of the activation pattern. Inverse-distance weighting never overshoots, but for powers above 1 its gradient vanishes at every point, leaving flat bull’s-eyes that make conduction look fast. Harmonic interpolation on the mesh solves Laplace’s equation between the points. It reproduces linear fields when the boundary is fixed, but treats a free edge as if nothing crossed it. Isochrones, lines of equal time, are \(c\,\Delta\tau\) apart, so crowding signals slow conduction. Interpolation cannot represent a line of block: it turns a jump into a ramp that reads as conduction at spacing over delay. Finally, a map is colored only within a fill threshold of some point, and the colored fraction depends on point density.
From the field
- How many points are enough. In simulations and 21 clinical atrial tachycardia maps, the sampling density beyond which re-interpolated maps improved little was about 1.0–1.5 points/cm² overall. It was higher for reentry than for focal activation, and higher in the complex left atrium (Williams et al. 2018).
- Isochrone crowding as a target. One registry defined deceleration zones as more than 3 isochrones within a 1-cm radius on sinus-rhythm maps annotated to the latest deflection. In 120 patients they colocalized with VT termination sites in 95% of cases, as the study reports (Aziz et al. 2019).
- Display settings change the map. Early descriptions of electroanatomic mapping already listed the triangle fill threshold and color range among display controls (Marchlinski et al. 1998). Ripple mapping, which shows each electrogram as a moving bar instead of an interpolated color, was designed to overcome limitations of isochronal maps (Koa-Wing et al. 2015).
Derivations
Symbols used throughout. \(T\) activation time; \(T_i\) at mapped point \(i\); \(h\) point spacing; \(\lambda_i\) barycentric weights; \(d_i\) distance to point \(i\); \(p\) IDW power; \(L\) cotangent Laplacian (m06); \(\tau\) isochrone time; \(\Delta\tau\) isochrone step; \(c\) conduction velocity; \(\Delta T\) delay across a block; \(\rho\) point density; \(r\) fill radius.
Linear interpolation
Inside a triangle, \(T(\mathbf x) = \sum_i \lambda_i(\mathbf x)\,T_i\) with barycentric weights. Along a line of points spaced \(h\):
\[ T(\mathbf x) = \sum_{i=1}^{3}\lambda_i(\mathbf x)\,T_i, \qquad \lvert T - I_h T\rvert \le \frac{h^2}{8}\max\lvert T''\rvert \tag{1}\]
Linear fields are reproduced exactly; error comes from curvature. A focal activation is most curved near its source. 5 mm from a source at 1.0 m/s (Heil et al. 2023), points 5 mm apart gave a largest error of 0.45 ms against a bound of 0.63 ms (lab part A).
Inverse-distance weighting
\[ T(\mathbf x) = \frac{\sum_i d_i^{-p}\,T_i}{\sum_i d_i^{-p}}, \qquad \min_i T_i \le T \le \max_i T_i, \qquad \nabla T(\mathbf x_i) = \mathbf 0 \text{ for } p > 1 \tag{2}\]
Shepard introduced the method (Shepard 1968); m30 uses it. Near a data point the point’s own weight dominates, so for \(p > 1\) the map is flat there: a bull’s-eye. In the lab, with 60 points, the median gradient near points was 0.41 ms/mm for \(p = 2\) and 0.02 ms/mm for \(p = 3\), against a true 1.0 ms/mm. A speed computed from such a gradient (m25) is inflated (lab part B).
Simulation
Assumptions
- Model: focal activation from one point in a uniform 40 × 40 mm sheet; points placed at random with a fixed seed; inverse-distance weighting on a 61 × 61 grid (Equation 2); isochrones by marching squares (Equation 4). Tested against
epml.interpolation. - Parameters: conduction velocity 1.0 m/s (Heil et al. 2023); number of points, power and isochrone step are adjustable.
- What it ignores: annotation and position errors (m22, m16–m19), curved surfaces, block and slow regions.
Harmonic interpolation on a mesh
\[ (L\,T)_j = 0 \text{ at unmapped vertices } j, \qquad T_i \text{ fixed at mapped vertices } i \tag{3}\]
This uses the cotangent Laplacian of m06 (eq-m06-cotan). The solution never exceeds the data range (maximum principle), and on a flat mesh with the boundary fixed it reproduces linear fields exactly. A free boundary, such as a valve or vein opening in a chamber mesh, acts as a zero-flux condition. The map then bends isochrones to meet the edge at right angles. In the lab, a plane wave at 1.0 ms/mm gradient read \(-0.01\) ms/mm along the edge normal with a free edge, against 1.00 with the edge fixed. Harmonic maps from 200 points had less than half the RMS error of IDW maps (lab part C).
Isochrones
On a mesh with linear interpolation, the isochrone at time \(\tau\) crosses edge \((i, j)\) at
\[ \mathbf x = \mathbf x_i + \frac{\tau - T_i}{T_j - T_i}(\mathbf x_j - \mathbf x_i), \qquad \text{spacing between isochrones } \Delta x = c\,\Delta\tau \tag{4}\]
Crowded isochrones mean slow conduction or block, and widely spaced ones fast conduction. Isochronal late activation maps use crowding to find deceleration zones (Aziz et al. 2019). The spacing also depends on the step chosen for display.
Block and coverage
Two points \(h\) apart on either side of a line of block, activated \(\Delta T\) apart, are joined by a ramp; a map of \(\rho\) randomly placed points colored within a fill radius \(r\) covers a fraction:
\[ c_{\text{apparent}} = \frac{h}{\Delta T}, \qquad f_{\text{covered}} = 1 - e^{-\rho\pi r^2} \tag{5}\]
With points 4 mm apart whose activation times differ by 60 ms, the ramp reads 0.07 m/s: apparent slow conduction where nothing conducts (lab part E). Denser points make the ramp steeper but never vertical. With 150 points on a 50 × 50 mm patch (6 per cm²) and a 3-mm fill radius, 82% of the patch is colored.
Exact
Assumptions
- Model: activation along a line at constant speed with a single jump at a line of block; points sampled evenly, offset by half the spacing; linear interpolation and IDW on the sampled values (Equation 1, Equation 2, Equation 5). Tested against
epml.interpolation. - Parameters: conduction velocity 1.0 m/s (Heil et al. 2023); spacing, detour delay and IDW power are illustrative and adjustable.
- What it ignores: annotation of double potentials at the block (m22), noise, and the second dimension.
Key takeaways
- Interpolation fills in what was not measured; its errors depend on the method and the activation pattern.
- A ramp across a block and a flat spot at a point are both artifacts of interpolation, not physiology.
Clinical correlate
This section describes published findings only. It is not guidance for patient care, and it has not yet been reviewed by an EP physician.
Where this shows up in EP. Every activation map shown during a procedure is interpolated. Sampling density needs depend on how complex activation is (Williams et al. 2018). Isochrone crowding has been used to identify deceleration zones (Aziz et al. 2019). Interpolation settings and display thresholds change what a map shows (Marchlinski et al. 1998), and displays that show the electrograms themselves were designed to overcome limitations of isochronal maps (Koa-Wing et al. 2015).
What the evidence is. A combined simulation and clinical study of sampling density (Williams et al. 2018), a prospective registry (Aziz et al. 2019), descriptions of mapping displays (Marchlinski et al. 1998; Koa-Wing et al. 2015), and interpolation theory (Shepard 1968).
What a reader should ask:
- Which interpolation method, and does it flatten around points? (Equation 1, Equation 2)
- Are open edges handled, or do isochrones bend to meet them? (Equation 3)
- Is a crowded region slow conduction or interpolation across block? (Equation 4, Equation 5)
- How dense are the points, and what fill threshold is set? (Equation 5) (Williams et al. 2018)
Visual summary
Schematic
Assumptions
- Schematic: not to scale, no quantitative claim. Labels follow Equation 1, Equation 4 and Equation 5.
How experts think
Before reading an activation map, ask:
- Where are the actual points? Show them. (Equation 5)
- Which interpolant, and what does it do near points and edges? (Equation 2, Equation 3)
- Is the activation pattern curved where points are sparse? (Equation 1)
- Is crowding real slowing, or a block smeared by interpolation? (Equation 4, Equation 5)
- Would a different fill threshold change the conclusion? (Equation 5)
One sentence to keep: between the points, the map shows the interpolant, not the heart.
Cheat sheet
| Quantity | Result |
|---|---|
| Linear (Equation 1) | \(\sum\lambda_i T_i\); error \(\le h^2\max\lvert T''\rvert/8\) |
| IDW (Equation 2) | \(\sum d_i^{-p}T_i/\sum d_i^{-p}\); bounded; flat at points for \(p > 1\) |
| Harmonic (Equation 3) | \(L T = 0\) between points; free edges act as zero flux |
| Isochrones (Equation 4) | edge crossing \((\tau - T_i)/(T_j - T_i)\); spacing \(c\,\Delta\tau\) |
| Block, coverage (Equation 5) | apparent \(h/\Delta T\); covered \(1 - e^{-\rho\pi r^2}\) |
Values used in this module
| Value | Source |
|---|---|
| Optimal sampling about 1.0–1.5 points/cm² | Williams et al. (2018) |
| Deceleration zone: more than 3 isochrones within 1 cm; colocalized in 95% (120 patients) | Aziz et al. (2019) |
| Conduction velocity 1.0 m/s | Heil et al. (2023) |
| Point counts, delays, fill radii, source positions | illustrative |
| Errors, gradients, apparent speeds, coverage | our own code |
Interview questions
Try each question before you open its model answer. In a real interview, say which assumptions you are making.
Q1. What does a mapping system show between mapped points?
Model answer
An interpolated value; the choice of interpolant shapes the map (Equation 1, Equation 2, Equation 3).
Q2. Bound the error of linear interpolation along a line.
Model answer
\(h^2\max\lvert T''\rvert/8\); it is exact for linear fields and worst where activation curves, near a focal source (Equation 1).
Q3. Why does IDW produce bull’s-eyes?
Model answer
For \(p > 1\) the nearest point’s weight dominates so strongly that the gradient vanishes at every data point (Equation 2) (Shepard 1968).
Q4. What does that do to conduction velocity maps?
Model answer
Speed is the inverse of the gradient magnitude, so flat spots read as very fast conduction (Equation 2); m25 covers estimators that avoid this.
Q5. What is harmonic interpolation, and what is its advantage?
Model answer
Solving \(L T = 0\) between mapped vertices on the mesh; it follows the surface, cannot overshoot, and reproduces linear fields with a fixed boundary (Equation 3).
Q6. What happens at an open edge of the mesh?
Model answer
It behaves as zero flux: isochrones bend to meet the edge at right angles, whatever the true activation (Equation 3).
Q7. How far apart are isochrones?
Model answer
\(c\,\Delta\tau\) for a front at speed \(c\) and step \(\Delta\tau\) (Equation 4).
Q8. A map shows crowded isochrones. Name two explanations.
Model answer
Slow conduction, or interpolation across a line of block, which turns a jump into a steep ramp (Equation 5).
Q9. Points 4 mm apart straddle a block, and their activation times differ by 60 ms. What speed does a linear map show?
Model answer
4/60 ≈ 0.07 mm/ms, i.e. 0.07 m/s (Equation 5).
Q10. How many points does a map need?
Model answer
It depends on activation complexity and geometry; one study found about 1.0–1.5 points/cm² beyond which maps improved little (Williams et al. 2018). m26 goes further.
Q11. What fraction of a surface is colored for random points of density ρ and fill radius r?
Model answer
\(1 - e^{-\rho\pi r^2}\) (Equation 5).
Q12. How were deceleration zones defined in one VT study?
Model answer
More than 3 isochrones within a 1-cm radius on isochronal late activation maps (Aziz et al. 2019).
Code lab
The lab interpolates synthetic activation.
| Part | What you do | Data |
|---|---|---|
| A | Linear interpolation and its error (Equation 1) | synthetic |
| B | Inverse-distance weighting (Equation 2) | synthetic |
| C | Harmonic interpolation on a mesh (Equation 3) | synthetic |
| D | Isochrones (Equation 4) | synthetic |
| E | Block and coverage (Equation 5) | synthetic |
- Python notebook: labs/python/m24_lab.ipynb. Run it from the course repository after
pip install -r requirements.txt, or open it in Google Colab. - Data: none to download.
Each part ends with checks: lines that pass with the code as given. If you change something and a check fails, that is information, not an error to hide.
Self-check
Short checks you can do in your head. Open each answer after you have tried it.
S1. Halve the point spacing. What happens to the linear-interpolation error bound?
Answer
It falls to a quarter (Equation 1).
S2. Three points with times 10, 20, 30 ms. Can IDW return 35 ms anywhere?
Answer
No; it stays within 10–30 ms (Equation 2).
S3. Speed 0.5 m/s, isochrone step 10 ms. Isochrone spacing?
Answer
5 mm (Equation 4).
S4. An edge from 12 ms to 20 ms. Where does the 14-ms isochrone cross it?
Answer
A quarter of the way along (Equation 4).
S5. Points 2 mm apart across a block, activation times 100 ms apart. Apparent speed?
Answer
0.02 m/s (Equation 5).
S6. One point per cm², fill radius 5 mm. Covered fraction?
Answer
\(1 - e^{-\pi/4} \approx\) 0.54 (Equation 5).
Literature figures
These figures are worth looking at alongside this module. We link to them instead of copying them: some licenses do not allow reuse here, and others we could not confirm. Each entry says what to look for and where this module explains it.
- Re-interpolated maps at falling density. Williams and colleagues’ maps and error curves against sampling density (Williams et al. 2018). This module: block and coverage.
- Isochronal crowding. Aziz and colleagues’ isochronal late activation maps with deceleration zones (Aziz et al. 2019). This module: isochrones.
- Shepard’s surfaces. Shepard’s interpolated surfaces for different powers (Shepard 1968). This module: inverse- distance weighting.
Further reading
- Interpolation. Shepard’s method (Shepard 1968); LAT and conduction-velocity algorithms (Cantwell et al. 2015); the OpenEP platform (Williams et al. 2021).
- Sampling. Density for atrial tachycardia maps (Williams et al. 2018).
- Clinical use of isochrones. Deceleration zones (Aziz et al. 2019); electrograms shown as moving bars (Koa-Wing et al. 2015).
- Related modules. m06 (meshes), m22 (annotation), m25 (conduction velocity), m26 (map quality), m30 (capstone).