Fig 1.
Atrial model and localization of the ectopic foci.
Localization of the 58 ectopic focus sites (including the SAN) shown from six different views of the atria. Blue points labelled as R1 to R17 and L1 to L14 are the ectopic sites belonging to the validation set. Red points labelled as R18 to R30 and L15 to L27 represent the additional ectopic sites to cover the atrial walls.
Fig 2.
Simulated BSPMs and integral map (BSPiM) computed at the torso surface.
Torso model (including the atria and the additional regions) showing an example of the frontal (a) and rear (b) views of the Body Surface Potential Map (BSPM) at time t = 43 ms, applying the ectopic beat at position L1. The dotted square represents the approximate area of the frontal and rear surface where electrographic signals may provide useful information; (c) morphology of the P-wave (potential versus time) registered at the precordial lead V3 (highlighted as a white sphere in (a) in sinus rhythm (red line) and when stimulating ectopic L1 (blue line); d) Characteristic P-wave integral map (BSPiM) (left). Greenish to yellowish area represents positive P-waves registered at any lead of this area (see green P-wave line in panel (e) registered at the position with maximum integral value), and greyish to blueish region means negative P-waves (see purple P-wave line in panel (e) registered at the position with the minimum integral value); e) Dashed horizontal baseline represents the division between positive and negative potentials. Dotted vertical line defines the duration of the P-wave defined as the latest atrial depolarization time for each simulation (t = 143 ms in the case of L1).
Fig 3.
The atrial-torso model was built up and stabilized in sinus rhythm. Afterwards, 57 new biophysical simulations were carried out stimulating at 57 additional ectopic foci in order to compute the corresponding BSPMs and BSPiMs. Machine learning techniques were used to firstly cluster these ectopic sites into spatial and well-defined atrial regions and to secondly be able to prospectively classify and predict the origin of new sites with high accuracy.
Fig 4.
Comparison between simulated and experimental BSPiMs.
(a) Simulated Local Activation Times (LATs) computed when the ectopic sites are located at different positions of the RA (2, 5, 9, 12 and 17) as defined in [8] and locations of the LA (1, 4, 9, 10 and 14) as defined in [10,11]. Bluish colour corresponds to t = 0 ms and reddish colours correspond to the latest activation time. Time is given in milliseconds; (b) Simulated normalized BSPiM computed at the 14.157 nodes of the torso surface for each ectopic focus. Bluish colours correspond to the most negative while yellowish correspond to the most positive integral values. Red arrows represent the position of the minima towards maxima integral values; (c) Experimental integral patterns reproduced from the originals published by SippensGroenewegen et al in [8–11]. White colour means negative integrals (equivalent to the blueish range for the simulated BSPiM) while green colour means positive integrals (equivalent to the greenish to yellowish range for the simulated BSPiM).
Fig 5.
P-wave morphology registered in precordial leads produced by simulations (left panel) and by experimental activations (right panel).
Left panel: P-wave registered at V1 (blue), V3 (red) and V5 (green) from the RA simulations 15, 18, 25, 27 and 30 and from the LA simulations 2, 3, 6, 7, 9 and 10; Right panel: P-waves reproduced from the experimental recordings by Kistler et al in [14] at the same sites as used to simulate. Experimental stimulated regions: ostium and body of the CS (CS os and CS body), higher and lower CT (High and Low CT), lower region of the TA (Low TA), right and left atrial appendages (RAA and LAA), left and right and superior and inferior PVs (LSPV, RSPV, LIPV, RIPV).
Fig 6.
Simulated normalized body surface P-wave integral maps (BSPiM).
BSPiMs have been computed at the torso surface for the 26 additional ectopic foci stimulated at a) RA (R18 to R30), and b) LA (L15 to L27). Normalization was carried out based on the maximum range of P-wave integral considering the 58 different BSPiMs.
Fig 7.
Exemplary BSPiMs included in ECs for different number of clusters (k) from 2 to 6.
A single pattern belonging to each coloured class for each k is displayed.
Fig 8.
Clustering and classification of BSPiMs.
a) Clustering of the 57 patterns of BSPiM (30 from RA and 27 from LA) plus the SAN using the K-means algorithm on the torso surface nodes where K is the number of pre-defined ECs. For each K, all the foci belonging to the same EC have the same colour. b) For each K, left column shows the colour and number of each EC resulting from the clustering step (for example blue-k2-0 refers to blue EC, equivalent to class 0, when k = 2) while right column shows the classification results identifying the ectopic site that is not well classified with the colour of the correct group, to which it really belongs.