An ECG compression method exploiting a QRS detector for sparse dictionary learning

• Published in: Measurement : journal of the International Measurement Confederation Vol. 258; p. 119177
• Main Authors: Kovacova, Antonia, Iadarola, Grazia, Vito, Luca De, Kovac, Ondrej, Saliga, Jan, Sevec, Jergus
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 30.01.2026
• Subjects: Biomedical measurement system; Compressed sensing; DBBD; DCT; ECG; Internet of things; multiple measurement vector reconstruction (MMV); Signal recovery; Wearable devices; Internet of things; DCT; multiple measurement vector reconstruction (MMV); ECG; Signal recovery; Biomedical measurement system; Wearable devices; Compressed sensing; DBBD
• ISSN: 0263-2241
• DOI: 10.1016/j.measurement.2025.119177

•Compressed sensing using sparse dictionary learning with Discrete Cosine Transform (DCT) and MMV-OMP (Multiple Measurement Vector - Orthogonal Matching Pursuit) allows efficient ECG compression by using frames from one heart-depolarization cycle, aligned by the QRS complex.•A Pan-Tompkins detector, improved with multi-lead adjustment, is used to accurately find QRS complexes.•The Deterministic Binary Block Diagonal (DBBD) matrix is chosen as a sensing matrix since its combination with DCT functions provides good compression performance while keeping the design simple and low in complexity.•Compressed frames from each ECG recording are reconstructed simultaneously using MMV-OMP.•The method reaches a high compression ratio of up to 12, while keeping PRD (Percent Root-Mean-Square Difference) low, while diagnostic quality is well preserved, with WDD (Weighted Diagnostic Distortion ) results rated from good to very good, and WEDD (Wavelet Energy–based Diagnostic Distortion ) indicating very good to excellent reconstruction. This paper presents a Compressed Sensing (CS) method for electrocardiogram (ECG) using sparse dictionary learning for dimensionality reduction that exploits frames of one heart-depolarization cycle. The ECG signal is first acquired at the Nyquist rate and then segmented into multiple frames, with each frame aligned depending on the QRS complex positions detected by the Pan-Tompkins algorithm. During the training phase, a dictionary built through the Discrete Cosine Transform (DCT) is reduced through the Multiple Measurement Vector (MMV) algorithm. The compression employs the Deterministic Binary Block Diagonal (DBBD) matrix as a sensing matrix. The ECG frames are reconstructed by solving the MMV problem, and individual frames are aligned based on the R-peak value. This proposed method enables efficient data compression while preserving essential ECG signal information. The method achieves a high compression ratio of 12 while maintaining a low PRD, demonstrating its efficiency without compromising signal quality. Reconstruction quality was evaluated using both Weighted Diagnostic Distortion (WDD) and the Wavelet Energy–based Diagnostic Distortion (WEDD) metrics, showing very good to good WDD values up to CR = 12 and WEDD values indicating very good to excellent reconstruction.