A statistical approach for accurate detection of atrial fibrillation and flutter

We have developed an algorithm for real-time detection of atrial fibrillation (AF) and flutter (AFL) that relies on 4 statistical techniques (root mean square of successive differences, turning point ratio and Shannon entropy). For AFL detection we use a time-frequency based method via the complex demodulation technique to recognize a characteristic rhythm of AFL. We used the MIT-BIH Atrial Fibrillation database to train the algorithm and then tested it on the MIT-BIH Arrhythmia database. We achieved sensitivity (specificity) of 94% (95%) on the training set and 90 % (91%) on the test database for AF detection. We further tested the AF performance on 36 actual Holter files obtained from The ScottCare Corporation and achieved sensitivity and specificity of 95% and 95% respectively. AFL detection was tested on 2 files of the AFIB database and we achieved 97% sensitivity and 95% specificity. The ease of implementation and minimal memory and speed requirements make this algorithm ideal for on-chip or similar usage.