Session PB4.3

Image Processing on ECG Chart for ECG Signal Recovery

TW Shen*, TF Laio

Tzu Chi University
Hualien, Taiwan

Introduction: Medical imaging plays an indispensable role on medical informatics. Most of imaging processing technologies is focused on the identification of the locations of diseases on MRI, CT, PET, and SPECT images to assist the physicians on medical diagnosis. However, only a few researches focused on one dimension signal recovery or reconstruction of electronic signals from clinical paper charts to help biomedical engineers for developing computer aid diagnosis (CAD) tools.
Aim: The aim of this research is to recognize ECG phantom on charts, to recovery raw ECG signal, and to evaluate the performance of each imaging processing skills.
Methods: Spatial and frequency analysis were provided to process on color or gray-level electrocardiogram charts, including threshold segmentation and 2D Fourier transform methods. First, threshold segmentation or 2D Fourier transform were applied to remove the chart grids and marker lines. Second, column searching on missing pixels or dots was utilized, which were refilled by image matching and pixel interpolation. Then, the 2D image was converted to 1D signal by using five interpolation methods (cubic, v5cubic, pchip, spline, linear, and nearest) that rebuild sampling frequency to 500 sps. Finally, the performance was evaluated by percentage of root mean square (PRD) for calculating waveform similarity.
Results: MIT/BIH Sudden Cardiac Death Holter Database was investigated. Twenty-three ten-second charts (about 124 by 990 pixels) were processed and compared with original ECG data for performance evaluation. The spatial and frequency methods provided each best result on average 41.7% and 56.6% PRD by using linear interpolation.
Discussion: Overall, our image processing methods converted 2D ECG images into one-lead ECG signals to make further CAD analysis and modeling simulation possible. The results show that the spatial method offered better performance than the frequency method, especially if the RGB colors in chart are kept. Unlike threshold segmentation method, 2D Fourier transform method contains more salt-and-pepper noise on chart images and the recovered signals also include amplitude displacement. In addition, linear interpolation is the best interpolation method with less overall PRD.

(Abstract Control Number: 224)