Artificial intelligence-assisted gastric antral rhythm phenotyping and heart rate variability for assessment of gastrointestinal dysfunction after acute brain injury: a multicentre prospective cohort study

Gastrointestinal dysfunction (GID) is common after acute brain injury (ABI), but bedside assessment remains largely symptom based and provides limited information on gastric antral rhythm or autonomic regulation. We evaluated whether AI-assisted, curve-based gastric antral rhythm phenotypes were associated with daily gastrointestinal status assessed using GIDS and with heart rate variability (HRV)-derived measures reflecting aspects of autonomic regulation in patients with ABI. This multicentre prospective observational cohort study enrolled adult ICU patients with ABI from three teaching hospitals in Sichuan, China, between 15 September 2025 and 14 January 2026. On ICU days 1, 2, 3 and 7, a 6-min gastric antral ultrasound video was acquired during continuous ECG recording, and HRV was analysed over the synchronous 6-min interval. AI-based frame-by-frame segmentation generated gastric antral cross-sectional area (CSA) curves, which were independently classified as regular, irregular or gastroparesis-like by two trained investigators after quality control. HRV parameters were derived using Kubios HRV. Daily GID was assessed using the Gastrointestinal Dysfunction Score (GIDS), with GIDS ≥ 1 defining GID. Associations were analysed using patient-clustered generalised estimating equations. Ninety-eight patients and 271 monitoring days were analysed. GIDS ≥ 1 was observed on 75.6% of monitoring days. Regular, irregular and gastroparesis-like phenotypes accounted for 38.0%, 24.4% and 37.6% of monitoring days, respectively. Compared with the regular phenotype, the gastroparesis-like phenotype was associated with higher GIDS total score (adjusted β = 0.293, 95% CI 0.007–0.578; P = 0.045) and higher odds of GIDS ≥ 1 (adjusted OR = 1.991, 95% CI 1.070–3.704; P = 0.030). In unadjusted comparisons, the gastroparesis-like phenotype had a higher mean heart rate and lower SampEn, SDNN and RMSSD than the combined non-gastroparesis-like phenotype. After multivariable adjustment, the gastroparesis-like phenotype remained associated with higher mean heart rate, stress index and DFA α1 and lower SampEn and SDNN. In this multicentre prospective cohort of patients with ABI, AI-assisted gastric antral rhythm phenotypes derived from continuous ultrasound CSA curves were associated with daily GIDS and selected HRV-derived measures. These associations do not establish causality or diagnostic equivalence to clinical gastroparesis.