Diagnostic measure to quantify loss of clinical components in multi-lead electrocardiogram

Abstract

In this Letter, a novel principal component (PC)-based diagnostic measure (PCDM) is proposed to quantify loss of clinical components in the multi-lead electrocardiogram (MECG) signals. The analysis of MECG shows that, the clinical components are captured in few PCs. The proposed diagnostic measure is defined as the sum of weighted percentage root mean square difference (PRD) between the PCs of original and processed MECG signals. The values of the weight depend on the clinical importance of PCs. The PCDM is tested over MECG enhancement and a novel MECG data reduction scheme. The proposed measure is compared with weighted diagnostic distortion, wavelet energy diagnostic distortion and PRD. The qualitative evaluation is performed using Spearman rank-order correlation coefficient (SROCC) and Pearson linear correlation coefficient. The simulation result demonstrates that the PCDM performs better to quantify loss of clinical components in MECG and shows a SROCC value of 0.9686 with subjective measure.

Keywords: MECG data reduction scheme; MECG enhancement; MECG signals; Pearson linear correlation coefficient; Spearman rank-order correlation coefficient; clinical components; electrocardiography; medical signal processing; multilead electrocardiogram signals; principal component analysis; principal component-based diagnostic measure; wavelet energy diagnostic distortion; weighted diagnostic distortion; weighted percentage root mean square difference.

Fig. 1

Evaluation of PCDM for MECG signals. PCA is the abbreviation for PCA

Fig. 2

ER against PCs for HC, MI and DT

Fig. 3

First six PCs of HC MECG and frequency responses of a–f First six PCs of HC MECG g Frequency response of PC1 h Frequency response of PC2 i Frequency response of PC3 j Frequency response of PC4 k Frequency response of PC5 l Frequency response of PC6

Fig. 4

Block diagram of the proposed MECG data reduction technique

Fig. 5

Block diagram of MECG reconstruction from the coded data

Fig. 6

Noisy ECG signals and the filtered (processed) ECG signals of leads I, II, V1 and V2 a–d Noisy ECG signal with SNR of 1 dB at leads I, II, V1 and V6 e–h Filtered ECG signal at leads I, II, V1 and V6

Fig. 7

Original and processed ECG signals at leads I, II, V1 and V6 a–d Original ECG signal at leads I, II, V1 and V6 e–h Processed ECG signal at leads I, II, V1 and V6

Fig. 8

Variation of a Variation of CR with number of quantisation bits b Variation of MPRD, MWEDD and proposed PCDM diagnostic measures with respect to number of quantisation bits c Variation of diagnostic measures such as MPRD, MWEDD and PCDM with respect to CR d DAOS versus PCDM graph for MECG signals