The discrimination between atrial fibrillation (AFib) and regular atrial arrhythmias including atrial flutter (AFlu) and focal atrial tachycardia poses a diagnostic challenge both to physicians and computerized algorithms.
As a result, misinterpretation rates of up to 80% have been reported. AFib represents high-frequent chaotic electrical activation of the atria exhibiting ECG signs of fibrillation waves in combination with a chaotic ventricular response.
In contrast, electrical activation follows defined reentrant circuits in AFlu resulting in regular flutter waves. However, the discrimination of AFib and AFlu from the surface ECG can be made very difficult by several factors. AFib may present with coarse fibrillatory waves, reminiscent of AFlu. AFlu may display atypical characteristics including hardly discernible low voltage flutter waves as well as an irregular ventricular response. However, the exact differentiation between AFib and AFlu is imperative with respect to treatment modalities as the effectiveness of antiarrhythmics is generally lower in AFlu and catheter ablation is often the superior option. Furthermore, atypical forms of AFlu are becoming increasingly important in clinical practice as a complication of left atrial ablation procedures.
Many mathematical models have been proposed to represent electrical conductivity in the heart, most prominently the Noble adaptation of differential equations, for which Hodgkin and Huxley were distinguished with the Nobel Prize in 1963. In this mathematical model the electrical potential across the membrane changes due to ion currents and is related to sodium and potassium currents. This mathematical model successfully predicted several so far unknown phenomena and lead to many extensions.
In [S1] we proposed a mathematical multi-level model that is based on phenomenological observations. It is well known that there are two types of signal blocking behavior in the atrioventricular (AV) node: Mobitz and Wenckebach type blocks. The idea of multi-level AV block is to consider a sequence of filters of this easy type. We applied an optimization approach in the following sense: assuming a regular (AFlu) behavior, the objective function is the deviation from simulated signaling behavior from the real measurements (R wave peak times), with model parameters as degrees of freedom. When a low optimal objective function value can be obtained, we see this as an indication for AFlu, otherwise for AFib. For data from 100 patients, this approach resulted in a sensitivity of 89% and a specificity of 80% that can be compared to the 20% sensitivity reported in the literature or the 24% specificity that a statistical approach obtained in [S1] (worse than guessing!).