Clinical evaluation of algorithms for ST measurement during exercise test

Area of Science:

• Cardiology
• Biomedical Engineering
• Medical Informatics

Background:

• Computerized analysis of exercise electrocardiograms (ECGs) offers benefits but lacks standardized evaluation methods.
• Assessing the reliability of ECG analysis algorithms is challenging without annotated databases.
• Current performance evaluations rely on unannotated clinical recordings.

Purpose of the Study:

• To evaluate the accuracy of a novel microcomputer-based exercise test analyzer.
• To compare computer-derived ST segment measurements with those of experienced cardiologists.
• To establish the clinical reliability of automated ECG analysis.

Main Methods:

• Six hundred exercise ECG strips from 60 patients were randomly selected.
• Two cardiologists blindly assessed ST shift (at J + 80 ms) using calibrated lenses.
• Computer measurements were compared against observer assessments using correlation coefficients and regression analysis.

Main Results:

• Computer analysis had a minimal failure rate (0.83%) due to noise.
• Observer agreement was high, with few unreadable or rejected ECGs.
• Computer measurements showed statistically significant correlation (p < 0.001) with cardiologists' assessments, with no systematic bias.

Conclusions:

• The developed algorithms achieve accuracy comparable to trained cardiologists in measuring exercise ECG ST changes.
• Further research with dedicated, annotated databases is needed to assess computer analysis on uninterpretable ECGs.
• Computer-assisted analysis holds potential for improving ECG interpretation accuracy, especially in noisy signals.