Adaptive wavelets applied to automatic local activationwave detection in fractionated atrial electrograms of atrial fibrillation

Catheter ablation is an effective therapy to treat atrial fibrillation (AF) whenever the proper atrial regions are targeted. Electro-anatomical mapping is commonly used for that purpose, thus facilitating the location of ablation targets. However, reliable mappings acquisition depends on an accurate detection of local activation waves (LAWs) from atrial electrograms (EGMs). This is currently a handmade and time-consuming task performed during the intervention. In this work a novel algorithm to detect automatically LAWs is proposed. To deal with complex and fractionated recordings, the EGM is decomposed making use of a tailor-made wavelet function. Such a function is generated from the atrial activation providing the highest average correlation within the EGM. According to manual annotations provided by two experts from 21 EGMs, the algorithm identified 959 out of 970 available LAWs. Thus, for the whole database its average sensitivity, positive predictivity and accuracy were 99.18% ± 1.35%, 99.69% ± 0.66% and 98.90% ± 1.51%, respectively. These results suggest the method's reliability, being able to detect the LAWs and ignoring successfully non-atrial patterns, such as noise, artifacts or other baseline oscillations, which can often lead to false detections.