Search

An automatic, real-time localization of ventricular tachycardia (VT) can improve the efficiency and efficacy of interventional therapies. Because the exit site of VT gives rise to its QRS morphology on electrocardiograms (ECG), it has been shown feasible to predict VT exits from 12-lead ECGs. However, existing work have reported limited resolution and accuracy due to a critical challenge: the significant inter-subject heterogeneity in ECG data. In this paper, we present a method to explicitly separate and represent the factors of variation in data throughout a deep network using denoising autoencoder with contrastive regularization. We demonstrate the performance of this method on an ECG dataset collected from 39 patients and 1012 distinct sites of ventricular origins. An improvement in the accuracy of localizing the origin of activation is obtained in comparison to a traditional approach that uses prescribed QRS features for prediction, as well as the use of a standard autoencoder network without separating the factors of variations in ECG data.

Existing criteria recommended by ACC/ESC for identifying patients with ST-elevation myocardial infarction (STEMI) from the 12-lead ECG perform with high specificity (SP), but low sensitivity (SE). In our previous study, we found that the SE of acute ischemia detection can be markedly improved without any loss of SP by calculating, from the 12-lead ECG, ST elevation in 3 vessel-specific leads (VSLs). To further validate the method, we evaluated the SP using a dataset with non-ischemic ST-segment changes, consisting of 12-lead ECGs of 100 patients. These ECGs were chosen to represent five causes of pathological ST deviation, other than acute coronary occlusion: ventricular pre-excitation, acute pericarditis, early-repolarization syndrome, left ventricular hypertrophy, and left bundle branch block. Both STEMI and VSL criteria were tested by calculating SP as the performance measure. We found that SP of the STEMI criteria was 100%, 4%, 29%, 100%, and 64%, respectively, for the five subgroups. The corresponding values of SP for the VSLs were 92%, 88%, 100%, 77%, and 68%. For the entire group, SP was 57% for the STEMI criteria and significantly higher for the VSLs at 83%. Thus, the VSLs not only are more sensitive in detecting acute ischemia, but also significantly more specific in rejecting patients with non-ischemic ST deviation than the existing STEMI criteria.

It has been proposed that by optimizing the timing of activation between the ventricles (V-V interval), with the aid of body-surface potential mapping (BSPM), the success rate of cardiac resynchronization therapy (CRT) devices could be improved. We recorded 120-electrode BSPM data and CT scans from two patients with implanted CRT devices and calculated, by means of electrocardiographic inverse solution, epicardial potentials and isochrones of activation for different V-V intervals. To gain better insight, we used a computer model of ventricular activation to simulate activation isochrones for CRT pacing. As a measure of inter-ventricular synchrony for a variety of V-V settings we used the area between the LV and RV percent surface activated curves. We have demonstrated that by aiming to minimize dyssynchrony in ventricular activation patterns, an optimal CRT pacing V-V interval can be selected. The computer model simulations provided a baseline measure by which our measure of synchrony can be evaluated.

We hypothesized that the calculation of epicardial potentials from body-surface potential maps (BSPMs) could aid ablation of ventricular tachycardia (VT). BSPMs were recorded during epicardial catheter mapping and pacing in 2 patients. Single-beat epicardial maps were calculated by inverse solution using customized torso/cardiac geometry, discretized from a CT scan. During pacing from 48 epicardial sites, we observed stimulus-QRS delay (Stim-QRS) of 27 plusmn 7 ms and a difference between known pacing locations and calculated sites of earliest potential minima of 1.6 plusmn 1.4 cm. Pacing in scar and scar-border zones had longer Stim-QRS delay (51 plusmn 24 ms and 35 plusmn 23 ms, respectively, p=0.004), and greater distances between known pacing sites and known locations (3.0 plusmn 1.6 cm and 4.6 plusmn 2.0 cm, respectively, p=0.0004). BSPM with inverse solution mapping can identify sites of earliest epicardial activation and thus could have clinical utility.