Improving efficiency, confidence, and readability in ECG interpretation; the role of signal processing

Abstract

Abstract Funding Acknowledgements Type of funding sources: Private company. Main funding source(s): Project privately funded by B-Secur Ltd Background Electrocardiogram (ECG) interpretation is a crucial component in the cardiac clinical pathway that enables the triaging, diagnosis, and treatment of numerous medical conditions. However, noise contamination during signal acquisition can hinder clinical interpretation by hiding or distorting the pathology of the underlying ECG. By removing noise to uncover the true beat morphology, signal processing algorithms maximize the diagnostic value of the acquired ECG and reduce clinical burden by eliminating the need for costly and time-consuming repeat investigations. Purpose To evaluate the impact of robust signal processing software on the clinical interpretation and diagnosis of lead I ECGs. Additionally, to investigate whether signal processing could streamline the diagnostic process by revealing pathologies not previously seen in the ECG, improving the speed of diagnosis, and enhancing the readability of the ECG. Methods ECG strips (42 x 10 s) representative of a variety of beat morphologies, arrhythmias, and noise types were taken from a proprietary database and provided to healthcare professionals (N = 35) pre- and post-signal processing. Participants possessed varying degrees of experience (12 x Junior (<5 years), 10 x Experienced (5-10 years), 13 x Senior (>10 years)) and specialization in cardiology. Participants were asked to provide a rhythm annotation and corresponding confidence score for each strip when provided in a blinded, randomized order. Participants then provided an additional rhythm annotation when the raw and processed signals were presented side by side, in addition to answering three subjective questions on the impact of signal processing on the ECG interpretation process. Results Across all participants, 80.2% of processed ECG strips were able to be given a confident diagnosis. This increased to 86% when pre- and post-processed signals were provided alongside one another – an improvement of 10.8% over the raw signals (75.2%). In response to the questions, participants agreed that the processed signal had a positive or very positive impact on; i) helping to reveal a pathology not previously seen in the raw ECG (57% of strips), ii) speeding up the diagnosis (71% of strips), and iii) improving the readability of the ECG (80% of strips). Conclusion By effectively removing noise whilst retaining true beat morphology, signal processing increased the amount of clinically diagnostic ECG data and allowed clinicians to make a more efficient, confident diagnosis. With the positive impacts observed across all participant experience levels, the results suggest that robust signal processing has the potential to streamline the clinical interpretation of ECGs.