Waite, Stephen

Abstract Body surface gastric mapping (BSGM) enables non-invasive measurement of gastric electrophysiology, but the signals are approximately 100 times weaker than cardiac potentials and overlap spectrally with motion artifacts, necessitating labor-intensive manual review that limits clinical scalability. We present an uncertainty-aware deep learning framework combining a signal reconstruction network with a parallel uncertainty estimation network to automate artifact correction in high-resolution BSGM. Models were trained on 2,398 multihour, 64-channel recordings from 27 international clinical sites using weak supervision, a physiology-aware loss function, and uncertainty-gated quality control. In an independent cohort of 127 patients, the system achieved relative reductions of 39% in signal reconstruction error, 9% in total data removed, and 23% in amplitude–movement correlation compared with the industry-standard Wiener filter. Improved signal fidelity altered automated clinical phenotyping in 7% of patients by recovering previously obscured gastric rhythms. Uncertainty-aware deep learning enables reliable automated artifact correction in body-surface gastric mapping, improving signal fidelity and enabling scalable clinical interpretation. The system is FDA-cleared (510(k) K252504) and deployed in clinical practice, demonstrating that data-driven artifact correction can meet regulatory requirements for medical devices and reduce dependence on specialist manual review.

Chronic nausea and vomiting syndromes (NVSs) are prevalent and debilitating disorders. Putative mechanisms include gastric neuromuscular disease and dysregulation of brain-gut interaction, but clinical tests for objectively defining gastric motor function are lacking. A medical device enabling noninvasive body surface gastric mapping (BSGM) was developed and applied to evaluate NVS pathophysiology. BSGM was performed in 43 patients with NVS and 43 matched controls using Gastric Alimetry (Alimetry), a conformable high-resolution array (8 × 8 electrodes; 20-mm interelectrode spacing), wearable reader, and validated symptom-logging app. Continuous measurement encompassed a fasting baseline (30 minutes), 482-kilocalorie meal, and 4-hour postprandial recording, followed by spectral and spatial biomarker analyses. Meal responses were impaired in NVS, with reduced amplitudes compared to controls (median, 23.3 microvolts versus 38.0 microvolts, P < 0.001), impaired fed-fasting power ratios (1.1 versus 1.6, P = 0.02), and disorganized slow waves (spatial frequency stability, 13.6 versus 49.5; P < 0.001). Two distinct NVS subgroups were evident with indistinguishable symptoms (all P > 0.05). Most patients (62%) had normal BSGM studies with increased psychological comorbidities (43.5% versus 7.7%; P = 0.03) and anxiety scores (median, 16.5 versus 13.0; P = 0.035). A smaller subgroup (31%) had markedly abnormal BSGM, with biomarkers correlating with symptoms (nausea, pain, excessive fullness, early satiety, and bloating; all r > 0.35, P < 0.05). Patients with NVS share overlapping symptoms but comprise distinct underlying phenotypes as revealed by a BSGM device. These phenotypes correlate with symptoms, which should inform clinical management and therapeutic trial design.