Computer-based techniques for the optimal extraction of medical data from graphical paper records

A system is described which uses a new technique for the conversion of two dimensional waveforms conventionally stored on pair to a one dimensional array of data values. Paper records containing the waveform data are first scanned using a high resolution, flatbed, black and white scanner. The two dimensional pixel based images that are captured are stored as image files in the computer. Three digital image processing techniques are applied to these stored images. The first is a heuristic method for the automatic removal of random noise and the filling in of gaps in the captured image which are errors introduced during the scanning process. The second technique involves thinning each pixel based waveform to derive its skeleton. In the final part of the processing, data values are extracted from the thinned waveforms. These values are then filtered to eliminate quantisation errors introduced in the scanning process. Results using ECG waveforms are presented