The use of artificial intelligence (AI), such as neural networks or other machine learning models, may help to bring down the manual ECG review burden in population screening for AF.
One challenge in automated ECG analysis is to discriminate between noise and very irregular rhythm. To avoid false negatives (recordings with AF are rejected), the difficult noisy recordings are included in the review group, and thus increasing the manual review burden.
In a recent study, a CNN (Convolutional Neural Network) was trained to discriminate between correct beat detections and false beat detections in noise. The algorithm was trained on data from handheld ECG devices and applied after ECG analysis performed by the Cardiolund ECG Parser.
The results showed that the number of false detections of atrial fibrillation could be reduced by 22.5% with a minor cost in sensitivity.
Cardiolund is working to bring the technology used in this research into our future products.
The study is reported in the publication: Identification of transient noise to reduce false detections in screening for atrial fibrillation by Halvaei H, Svennberg E, Sörnmo L, Stridh M, in Frontiers in Physiology 2021 Jun 4, Vol. 12:672875. PMID: 34149452; PMCID: PMC8212862. DOI: 10.3389/fphys.2021.672875