Automated detection of atrial fibrillation based on vocal features analysis

Area of Science:

• Cardiology
• Medical Technology
• Signal Processing

Background:

• Early detection of atrial fibrillation (AF) is crucial but challenging due to its often asymptomatic nature.
• Current screening methods for AF are limited and typically require skin contact with devices like electrocardiography.
• Voice signal analysis presents a novel, non-contact approach for medical diagnostics.

Purpose of the Study:

• To evaluate the efficacy of vocal feature analysis for detecting atrial fibrillation (AF).
• To assess the reliability and accuracy of a developed "AF indicator" derived from speech signals.

Main Methods:

• A prospective study involving patients with persistent AF undergoing cardioversion across two medical centers.
• Recording of sustained vowel phonations ("Ahh" and "Ohh") synchronously with ECG.
• Development of an algorithm to generate an "AF indicator" from speech analysis.

Main Results:

• The study analyzed 143 ("Ahh") and 142 ("Ohh") patients, with a mean age of 71.4 years and 43% females.
• The "AF indicator" demonstrated reliability, with significantly lower and more homogenous values in sinus rhythm post-cardioversion.
• High diagnostic performance was observed: Area Under the Curve >0.98 ("Ahh") and >0.89 ("Ohh"), with 95% sensitivity and 82% specificity for AF detection.

Conclusions:

• This study first demonstrates the feasibility and reliability of using voice analysis for AF identification with acceptable accuracy.
• The developed "AF indicator" shows higher accuracy when analyzing the "Ahh" vowel compared to "Ohh."
• Voice analysis offers a promising, non-invasive tool for AF detection, complementing existing screening methods.