Tensor-based detection of T wave alternans using ECG.

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• T wave alternans (TWA), characterized by ABABAB amplitude changes in ECG T waves, is linked to heart disease and sudden cardiac death risk.
• Robust automatic detection of TWA is challenging due to its low amplitude in ECG signals.

Purpose of the Study:

• To develop and validate a novel, robust method for detecting T wave alternans in multichannel ECG signals.
• To leverage tensor decomposition for improved TWA detection accuracy, especially in noisy ECG segments.

Main Methods:

• A novel method using tensors (multidimensional matrices) to combine information from multiple ECG channels.
• Application of Canonical Polyadic Decomposition to tensors constructed from T waves over a 128-beat sliding window.
• Sign change counting on resulting loading vectors to determine TWA length and magnitude.

Main Results:

• The tensor-based method demonstrated enhanced reliability in detecting T wave alternans.
• Patients with positive clinical TWA tests showed significantly larger TWA length and magnitude compared to controls.
• The method effectively utilizes incomplete tensor decomposition for handling noisy ECG data.

Conclusions:

• The proposed tensor-based approach offers a more reliable method for T wave alternans detection in multichannel ECG.
• This technique shows potential for improved risk stratification of sudden cardiac death.
• The method's ability to handle noisy data enhances its clinical applicability.