Detection and removal of pacing artifacts prior to automated analysis of 12-lead ECG

A novel framework for the removal of pacing artifacts from bio-electrical recordings

BACKGROUND

Electroceuticals provide clinical solutions for a range of disorders including Parkinson's disease, cardiac arrythmias and are emerging as a potential treatment option for gastrointestinal disorders. However, pre-clinical investigations are challenged by the large stimulation artifacts registered in bio-electrical recordings.

METHOD

A generalized framework capable of isolating and suppressing stimulation artifacts with minimal intervention was developed. Stimulation artifacts with different pulse-parameters in synthetic and experimental cardiac and gastrointestinal signals were detected using a Hampel filter and reconstructed using 3 methods: i) autoregression, ii) weighted mean, and iii) linear interpolation.

RESULTS

Synthetic stimulation artifacts with amplitudes of 2 mV and 4 mV and pulse-widths of 50 ms, 100 ms, and 200 ms were successfully isolated and the artifact window size remained uninfluenced by the pulse-amplitude, but was influenced by pulse-width (e.g., the autoregression method resulted in an identical Root Mean Square Error (RMSE) of 1.64 mV for artifacts with 200 ms pulse-width and both 2 mV and 4 mV amplitudes). The performance of autoregression (RMSE = 1.45 ± 0.16 mV) and linear interpolation (RMSE = 1.22 ± 0.14 mV) methods were comparable and better than weighted mean (RMSE = 5.54 ± 0.56 mV) for synthetic data. However, for experimental recordings, artifact removal by autoregression was superior to both linear interpolation and weighted mean approaches in gastric, small intestinal and cardiac recordings.

CONCLUSIONS

A novel signal processing framework enabled efficient analysis of bio-electrical recordings with stimulation artifacts. This will allow the bio-electrical events induced by stimulation protocols to be efficiently and systematically evaluated, resulting in improved stimulation therapies.

Automatic Identification of Involuntary Muscle Activity in Subacute Patients with Upper Motor Neuron Lesion at Rest-A Validation Study

Sustained involuntary muscle activity (IMA) is a highly disabling phenomenon that arises in the acute phase of an upper motor neuron lesion (UMNL). Wearable probes for long-lasting surface EMG (sEMG) recordings have been recently recommended to detect IMA insurgence and to quantify its evolution over time, in conjunction with a complex algorithm for IMA automatic identification and classification. In this study, we computed sensitivity (Se), specificity (Sp), and overall accuracy (Acc) of this algorithm by comparing it with the classification provided by two expert assessors. Based on sample size estimation, 6020 10 s-long sEMG epochs were classified by both the algorithm and the assessors. Epochs were randomly extracted from long-lasting sEMG signals collected in-field from 14 biceps brachii (BB) muscles of 10 patients (5F, age range 50-71 years) hospitalized in an acute rehabilitation ward following a stroke or a post-anoxic coma and complete upper limb (UL) paralysis. Among the 14 BB muscles assessed, Se was 85.6% (83.6-87.4%); Sp was 89.7% (88.6-90.7%), and overall Acc was 88.5% (87.6-89.4%) and ranged between 78.6% and 98.7%. The presence of IMA was detected correctly in all patients. These results support the algorithm's use for in-field IMA assessment based on data acquired with wearable sensors. The assessment and monitoring of IMA in acute and subacute patients with UMNL could improve the quality of care needed by triggering early treatments to lessen long-term complications.

Demographic and Methodological Heterogeneity in Electrocardiogram Signals From Guinea Pigs

Electrocardiograms (ECG) are universally used to measure the electrical activity of the heart; however, variations in recording techniques and/or subject demographics can affect ECG interpretation. In this study, we investigated variables that are likely to influence ECG metric measurements in cardiovascular research, including recording technique, use of anesthesia, and animal model characteristics. Awake limb lead ECG recordings were collected in vivo from adult guinea pigs using a platform ECG system, while recordings in anesthetized animals were performed using both a platform and needle ECG system. We report significant heterogeneities in ECG metric values that are attributed to methodological differences (e.g., ECG lead configuration, ECG recording platform, presence or absence of anesthesia) that persist even within the same cohort of animals. Further, we report that variability in animal demographics is preserved in vivo ECG recordings—with animal age serving as a significant contributor, while sex-specific influences were less pronounced. Methodological approaches and subject demographics should be fully considered when interpreting ECG values in animal models, comparing datasets between studies, or developing artificial intelligence algorithms that utilize an ECG database.