Tissue artifact removal from respiratory signals based on empirical mode decomposition

Abstract

On-line measurement of respiration plays an important role in monitoring human physical activities. Such measurement commonly employs sensing belts secured around the rib cage and abdomen of the test object. Affected by the movement of body tissues, respiratory signals typically have a low signal-to-noise ratio. Removing tissue artifacts therefore is critical to ensuring effective respiration analysis. This paper presents a signal decomposition technique for tissue artifact removal from respiratory signals, based on the empirical mode decomposition (EMD). An algorithm based on the mutual information and power criteria was devised to automatically select appropriate intrinsic mode functions for tissue artifact removal and respiratory signal reconstruction. Performance of the EMD-algorithm was evaluated through simulations and real-life experiments (N = 105). Comparison with low-pass filtering that has been conventionally applied confirmed the effectiveness of the technique in tissue artifacts removal.

Figure 1

Schematic of the data acquisition system.

Figure 2

An example of the mutual information criterion.

Figure 3

Respiratory component and synthetic tissue artifacts. (a) The “clean” respiratory component; (b) Synthetic tissue artifact of SNR = 4 dB, f_TA = 0.40 Hz; (c) Synthetic tissue artifact of SNR = 10 dB, f_TA = 0.55 Hz; (d) Synthetic tissue artifact of SNR = 16 dB, f_TA = 0.70 Hz; (e) Synthetic tissue artifact of SNR = 14 dB, f_TA = 1 Hz, θ_TA = 30°; (f) Synthetic tissue artifact of slow walking; (g) Synthetic tissue artifact of fast walking; (h) Synthetic tissue artifact of running.

Figure 4

Mutual information, mutual information ratio and power of the IMFs at SNR = 2 dB.

Figure 5

SER results from EMD and LPF methods in Case 1. Percentage values represent the improvement of EMD-based technique over LPF.

Figure 6

Examples of synthetic signals. Case 2: (a) SNR = 4 dB, f_TA = 0.40 Hz; (b) SNR = 10 dB, f_TA = 0.55 Hz; (c) SNR = 16 dB, f_TA = 0.70 Hz. Case 3: (d) SNR = 14 dB, f_TA = 1 Hz, θ_TA = 30°. Case 4: (e) slow walking; (f) fast walking; (g) running.

Figure 7

Mutual information, mutual information ratio and power of the IMFs in case 3.

Figure 8

Comparative results between “clean” respiratory signal and signal after tissue artifact removal. (Case 4 during fast walking). (a) Synthetic signal; (b) Comparative results with LPF; (c) Comparative results with EMD.