Optimizing non-invasive assessment of gastric activity: a fully automated channel-wise pipeline for multi-electrode electrogastrography

Abstract

Electrogastrography (EGG) is a non-invasive, and affordable technique for investigating gastric physiology. However, its broader adoption is hindered by low signal-to-noise ratios and the lack of standardized, objective procedures for channel-wise artifact rejection and signal quality assessment. To address these limitations, we developed a fully automated, data-driven pipeline (Gastric Signal Processing Toolbox, GSPT) for channel-wise EGG preprocessing and analysis, designed to preserve physiologically-relevant information even under high-noise conditions. GSPT exploits high-resolution time-frequency representations, to automatically identify and remove motion-related and electronic artifacts through entropy-based analysis of the signal's instantaneous energy and robust outlier detection through integration of multiple anomaly-detection strategies. The resulting signals are characterized through a channel-wise gastric fingerprint, comprising established EGG measures, such as dominant frequency, together with novel descriptors such as Renyi entropy. Channel quality relays on Monte Carlo surrogate analysis, while recording-level quality is evaluated through dominant frequency consistency across channels. GSPT further provides ridge-based estimates of instantaneous amplitude and phase and quantifies inter-channel interactions through Total Correlation, enabling the construction of a global gastric fingerprint. Validation on datasets containing controlled motion artifacts demonstrated robust signal recovery with improved signal-to-noise ratios at the dominant frequency. Artifact identification, assessed using precision and recall, was comparable or better than existing approaches. Gastric fingerprints showed high stability within subjects and across recording sessions. As a freely available, one-click MATLAB toolbox, GSPT reduces manual bias and standardizes EGG preprocessing and analysis, even in low-density recordings, potentially providing a reproducible framework for investigating gastric functions in psychological, neurological and pathological conditions.