Digital recovery of biomedical signals from binary images

  • Authors:

  • M. Sanromán-Junquera;I. Mora-Jiménez;A. J. Caamaño;J. Almendral;F. Atienza;L. Castilla;A. García-Alberola;J. L. Rojo-Álvarez

  • Affiliations:

  • Signal Theory and Communications Department, G-123, University Rey Juan Carlos, Camino del Molino s/n, Fuenlabrada-28943, Madrid, Spain;Signal Theory and Communications Department, G-123, University Rey Juan Carlos, Camino del Molino s/n, Fuenlabrada-28943, Madrid, Spain;Signal Theory and Communications Department, G-123, University Rey Juan Carlos, Camino del Molino s/n, Fuenlabrada-28943, Madrid, Spain;Cardiology Department, Hospital General Universitario Gregorio Marañon, Madrid, Spain and Electrophysiology Unit, Grupo Hospitales de Madrid, Spain;Cardiology Department, Hospital General Universitario Gregorio Marañon, Madrid, Spain;Cardiology Department, Hospital General Universitario Gregorio Marañon, Madrid, Spain;Arrhythmia Unit, Hospital Virgen de la Arrixaca de Murcia, Spain;Signal Theory and Communications Department, G-123, University Rey Juan Carlos, Camino del Molino s/n, Fuenlabrada-28943, Madrid, Spain

  • Venue:
  • Signal Processing
  • Year:
  • 2012

Quantified Score

Hi-index 0.08

Visualization

Abstract

Given the vast amount of historical clinical data to be incorporated from old hospital information systems into new emerging digital storing standards, digital recovery of paper-written one-dimensional biomedical signals is a relevant application. Signal recovery from noisy, black and white, grid paper printout recordings, is a real situation that has received little attention in the literature. In this paper we propose an integral, automatic approach, based on digital image processing principles, and implemented in four stages: (1) orientation correction of the scanned image, using the eigenvector decomposition of the foreground pixel coordinates, hence reducing the computational cost of subsequent Hough Transform; (2) grid detection, using the Discrete Cosine Transform on horizontal and vertical histogram projections; (3) signal waveform identification, using morphological operators; (4) conversion from the waveform in the image plane to the one-dimensional biomedical signal. Time synchronization between the digitized gold standard and the recovered signals, which is essential for performance evaluation, is addressed by using of contrast filters to extract fiducial points on both signals, which are then fitted to a regression curve. Results with black and white paper printout recordings of intracardiac signals show that proposed approach is capable of automatically recovering biomedical signals from noisy images.