Spatial shape variability of ECG waves

International audienceAn increased beat to beat variability of ECG waves shapes (mainly QRS), along the time, has yet been observed in patients with some pathological cardiac conditions. The aim here is to present a method to measure spatial shape variability of ECG waves recorded, at a fixed time, from a Body Surface Potential Mapping (BSPM) device with 64 channels. Applications to P, QRS and T waves were made and evaluated on the ability to separate MI patients from healthy subjects. The data come from averaged segments of HR ECG beats. For each column of leads, all the differences in shape between each couple of signals (P, QRS or T) associated to each couple of leads were computed using the Distribution Function Method (DFM). After selecting a reference signal in each column, all the computed shape distances were put in a vector of 61 components (associated to the recording leads), for each person. Principal Component Analysis (PCA) was applied to the data, using the Karhunen Loève Transform (KLT) to extract, in each group, a reduced basis of eigenvectors. First the eigenvalues of the covariance matrix of the data vectors of each group are computed. Then the basis is formed by the eigenvectors associated to the highest eigenvalues. It was observed that one eigenvalue was highly larger than the others. So each subject data vector was projected both on the eigenvector from the healthy group and on the one of the patient group. Finally, the numbers used for clustering were the differences between the two scalar products. The results show a very good separation of the groups according to T wave shape analysis (specificity 100%, sensitivity 100%), a quite good according to QRS (specificity 92%, sensitivity 100%). For the P wave shapes specificity is lower (specificity 69%, sensitivity 92%). The results are in agreement with the MI diagnosis for the patients