Signal Processing and Machine Learning Automated Evaluation of Phrenic Nerve Affectation by Cardiac Stimulation.

Cryo-ablation is a common procedure used in hospitals to eliminate certain arrhythmia, such as Atrial Fibrillation. This procedure, sometimes, involves treatments in areas close to the phrenic nerve with the subsequent risk of later damage to the aforementioned nerve. To avoid this, clinical practice incorporates manual safety protocols during ablation. We propose the development of an automated classifier that facilitates the clinical evaluation of possible conduction disorders produced in the phrenic nerve. To achieve this goal, polygraph signals extracted during the ablation process of ten patients were used. To unmask the residue of cellular muscle potential during the phrenic nerve stimulation process we compare utilizing signal processing the results when the sensor was placed on the phrenic nerve (activation capture) and when the sensor was displaced from the phrenic nerve (no capture). A linear classifier was applied to both situations to characterize muscle activity resulting from nerve activation. The results confirmed that it is possible to automatically classify the level of muscle activity from the phrenic nerve with 100% accuracy in this data set. The method proposed in this work constitutes an automated protocol to evaluate the eventual deterioration of the phrenic nerve conduction due to ablation in the vicinity, improving the existing protocol for clinical convenience.