Click here for full text:
Detection of Spontaneous Termination of Atrial Fibrillation
Logan, Beth; Healey, Jennifer
HPL-2004-165
Keyword(s): atrial fibrillation; machine learning
Abstract: We present techniques to detect various types of terminating and non-terminating atrial fibrillation (AF) as required by the Computers in Cardiology Challenge 2004. First, we describe an automatic technique to distinguish non-terminating AF from terminating AF. Our method models R-R intervals using mixtures of Gaussians and achieves an accuracy of 90% on the training set and 77% on the challenge test set. Second, we describe a semi-automatic technique to distinguish immediately terminating AF from AF which terminates one minute later. Our method first uses spectral models to determine which pairs of records are recorded from the same patient. This technique achieves 100% accuracy on the training set and partitions the test set into 10 unique record pairs. We then examine by hand the end of each ECG record to determine the likely time ordering of the records in each pair, thus distinguishing which record terminates immediately. This technique achieves an accuracy of 90% on the challenge test set. Notes: Copyright IEEE. Published in and presented at IEEE Computers in Cardiology, 19-22 September 2004, Chicago, IL
8 Pages
Back to Index
| 
Printable version
