Lightweight Electrocardiogram Signal Quality-Aware VT/VF Detector for Resource-Constrained Life-Threatening Monitoring Devices

• Published in: IEEE sensors letters Vol. 9; no. 7; pp. 1 - 4
• Main Authors: Phukan, Nabasmita, Manikandan, M. Sabarimalai, Pachori, Ram Bilas, Garg, Niranjan
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.07.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Arrhythmia; Cardiac arrhythmia; Constraints; Datasets; Electrocardiography; Lead; Monitoring; Noise; Quality assessment; Real time; Real-time systems; Recording; resource-constrained devices; Rhythm; Sensor applications; Signal quality; signal quality assessment (SQA); Tachycardia; Ventricular fibrillation; ventricular tachycardia (VT)/ventricular fibrillation (VF) detection; zero-crossing rate (ZCR)
• ISSN: 2475-1472; 2475-1472
• DOI: 10.1109/LSENS.2025.3570346

Ventricular tachycardia (VT) and ventricular fibrillation (VF) are life-threatening arrhythmias, which lead to sudden cardiac arrest (SCA). The timely detection of VT and VF is vital, as automated external defibrillators rely on accurate VT/VF identification to deliver life-saving defibrillation and restore normal sinus rhythm during SCA. Continuous monitoring of electrocardiogram (ECG) signals plays a pivotal role in the early detection of VT/VF, potentially reducing mortality associated with SCA. However, the reliability of continuous ECG monitoring is often compromised by various noise sources, necessitating assessment of signal quality to ensure accurate VT/VF detection. This letter presents a real-time signal quality assessment (SQA)-based VT/VF detection method using zero-crossing rate. The SQA-based VT/VF detection method is tested on single and multilead datasets. The method is tested on real-time ECG signals collected from subjects with cardiac arrhythmias. Compared to zero-crossing rate-based VT/VF detection without SQA, the proposed SQA-based method reduced the false detection rate by up to 7.38% on a single-lead dataset and 59.22% on lead 1 of a multilead dataset. The method, implemented on the Arduino Due, consumed energy of 5.79 mJ and processing time of 13 ms, validating its real-time feasibility on resource-constrained wearable health monitoring devices.