Diagnostic Interpretation of Non-Uniformly Sampled Electrocardiogram

Area of Science:

• Biomedical Engineering
• Signal Processing
• Cardiology

Background:

• The daily global volume of electrocardiogram (ECG) data is immense (approx. 600 TB).
• Long-term storage (approx. 40 years) necessitates efficient data representation.
• Current diagnostic methods may not be optimized for non-uniformly sampled ECG data.

Purpose of the Study:

• To develop and validate fundamental methods for ECG diagnostic interpretation of non-uniformly sampled signals.
• To assess the feasibility of medical-grade diagnostics using storage-efficient ECG representations.
• To enable accurate analysis of ECG data despite arbitrary sampling.

Main Methods:

• A refined time-independent QRS detection method using moving shape matching.
• A graph data representation for quantifying asynchronous heartbeat similarity.
• A correlation-based non-uniform to time-scale transform for multiresolution ECG representation and wave delimitation.

Main Results:

• QRS detection accuracy of 99.93% with a 0.18% false detection ratio.
• Classification accuracy of 99.27% for common ECG beat types (MIT-BIH) and 99.37% (AAMI labels).
• Wave delineation errors within acceptable limits for medical-grade software (e.g., 3.4 ms for QRS duration).

Conclusions:

• The proposed methods are effective for ECG diagnostic interpretation of non-uniformly sampled signals.
• Medical-grade diagnostics are feasible using storage-efficient, non-uniform ECG representations.
• The processing chain demonstrates high accuracy and acceptable error margins for clinical application.