Detection of atrial fibrillation in patients after stroke

[...]we wish to caution that a classification relying on detected atrial fibrillation is limited by the heterogeneous monitoring approaches for the assessment of subclinical atrial fibrillation and, even more so, by the fleeting nature of the often infrequent and asymptomatic paroxysmal atrial fibrillation. [...]we propose a fourth category that would include patients without previously known atrial fibrillation but who have clinical markers, such as an artificial intelligence (AI) enhanced electrocardiogram (ECG) indicating a high probability of subclinical or silent atrial fibrillation. Deep learning AI models based on convolutional neural network methods can be applied to raw ECG data to detect or predict various cardiac conditions.2 These AI-ECG algorithms do not rely on specific or pre-defined ECG characteristics and can detect pathology even when the ECG is seemingly normal by human expert interpretation. In an analysis of patients with embolic stroke of undetermined source, the AI-ECG-detected probability of atrial fibrillation correlated with the likelihood of atrial fibrillation detection by ambulatory monitoring.3 Patients with embolic stroke of undetermined source and a high AI-ECG probability of atrial fibrillation might even be considered for oral anticoagulation in the absence of an elevated bleeding risk.