New Developments in Creation, Preservation and Reuse of ECG Knowledge Bases

Suave Lobodzinski^a, Ulrich Teppner^b, Michael Laks^c

^aCalifornia State University, Long Beach, CA, USA
^bTechnische Hohschule, Berlin, Germany
^cHarbor-UCLA Medical Center, Torrance, CA, USA

Correspondence: S Lobodzinski, California State University, Dept. of Electrical and Biomedical Engieering, Long Beach, CA 90840, USA. E-mail: slobo@linkline.com, phone 562-985-5521, fax 562-985-5899.

1.    Introduction

The motivation for this research was to develop a method for building digital databases from hard copy repositories using the optical ECG waveform recognition method (OEWR) [1]. Most electrocardiograms collected over the years in both clinical trials and longitudinal population studies today are in paper form. Most vendors have used digital ECG formats for over 20 years. These were proprietary and non-compatible with each other. A multiplicity of vendor specific formats and lack of ECG data standardization prevented the inclusion of digital ECGs into electronic patient record and creation of vendor neutral ECG research databases. In large randomized trials, which can have populations in the tens of thousands [3], sending, analyzing, and archiving hard copy ECGs to central core ECG labs become a significant burden. Similar problems are encountered in large epidemiological studies (Framingham Heart Study), which are hampered by the slow pace of the manual ECG measurement process. Although large quantities of hard copy ECGs have been collected in on-going clinical trials these could not be stored in digital databases for future re-analysis and re-use.

      Figure 1. Harmonized ECG Knowledge Base.

2.    Materials and Methods

Hard copy ECGs records in 8.5 x 11 inch format were optically scanned (resolution 600 dpi / 0.042 mm/pixel, 24-bit RGB). The typical size of scanned ECG images was 100 Mbytes. The algorithm for OEWR written in Matlab 5.3 [1] comprised 3 steps: 1) ECG raster pre-filtering, 2) waveform segment linking and filtering, 3) waveform data extraction. A template was developed for the 4x3x1R ECG records printed on a laser printer by the Tracemaster ECG Management System (Hewlett-Packard) to facilitate both the text and waveform recognition. Text information containing patient demographics, annotations and study attributes data, was extracted from 8.5x11 printouts using optical character recognition (Lead Tools v.12.1). Both text and waveform data were displayed on a monitor for a visual check prior to XML conversion [6], [7] and submission to the database. The extracted ECG waveforms were further re-analyzed by the Hewlett-Packard Xli ECG Analysis program and both measurement and study attribute data were added to the records. Digital ECG records stored on the Tracemaster were up-sampled to 500 Hz and exported into the XML format. Both OEWR and Tracemaster ECG objects in the XML format were annotated by the expert cardiologists and inserted into Oracle 9i Release 2 database (Oracle Corporation, Redwood City, California). A block diagram of a harmonized ECG knowledge base (HEKB) is shown in Fig.1.

3.  Results

The OEWR extracted ECG waveforms showed average sample-by-sample absolute differences of 1 to 3 pixels (4 to 12mV) from the original digital data, with an exception of the R-wave upstroke region where due to limited sampling rate, the differences ranged from 10-20mV as shown (Fig. 2).

Figure 2. Overlaid Original and OEWR extracted ECG data points from leads I and aVR were overlaid and plotted alongside the sample-by-sample amplitude differences. Horizontal and vertical axes’ dimensions are in pixels - 1 second = 500 pixels and 1 milivolt = 200 pixels respectively.

Our implementation of HEKB fully supported W3C XML data model and provided standard access for navigating and querying XML ECG records. It allowed for direct storage of native XML encapsulated ECGs in the database, provided support for XML Schemas and XML output from SQL queries.

4.  Conclusions

Proposed method for creation of HEKB from hard copy ECG recordings is an efficient way to include legacy ECGs into research databases, thus, providing their preservation and accession [2]. The inclusion of historic records along with ancillary demographic and clinical information increases their research value [2]. Such databases can be used for genetic epidemiology studies [3], clinical outcomes, and development of statistical methodologies to be used with these data for developing and testing of ECG algorithms. Another important area for OEWR applications is cardiac safety monitoring in clinical trials.

Acknowledgements

This work was supported in part by NIH SBIR grant 1 R01 HL63629-02A2.

References

[1] Lobodzinski S; Teppner U; Kuzminska M; Laks M. Optical ECG Waveform Recognition. J Electrocardiol 2002 Jul;35(3):285  

[2] Norman JE; Bailey JJ; Berson AS; Haisty WK; Levy D; Macfarlane PM; Rautaharju PM NHLBI workshop on the utilization of ECG databases: preservation and use of existing ECG databases and development of future resources. Electrocardiol 1998 Apr; 31(2):83-915.

[3] Bressan M; Bortolan G; Cavaggion C; Fusaro S. Normal electrocardiogram in the aged (the ILSA (Italian Longitudinal Study of Aging) Project). G Ital Cardiol 1998 Jan;28(1):22-815.

[4] www.fda.gov/cder/regulatory/ersr/ECGdata.htm

[5] http://www.cdisc.org/