Objective Gastric hemorrhage is one of the most common and life-threatening emergencies of the upper digestive tract. Early diagnosis and continuous monitoring are critical for reducing the risk of rebleeding and mortality. Although conventional endoscopy and imaging can localize the bleeding site, their invasiveness and limited real-time capability restrict their clinical utility for continuous monitoring. This study proposes a three-dimensional electrical impedance tomography (3D-EIT)–based gastric hemorrhage detection method (3D-gEIT) aimed at achieving noninvasive, real-time, and dynamic monitoring of gastric bleeding. Methods A three-dimensional upper-abdominal model including the stomach was constructed, and three electrode configurations—dual-layer ring, four-layer staggered ring, and opposed dual-plane array—were designed to compare their imaging performance through simulation. A region-clustering constraint was incorporated into the Tikhonov–Noser hybrid regularization framework to develop the TK-Noser-RCC algorithm, which improves spatial coherence and noise robustness. An upper-abdominal physical phantom based on agar was then built, in which hemispherical inclusions of varying volumes (10–50 mL) were embedded to simulate different severities of bleeding. Boundary voltages were collected and 3D reconstructions were performed. In addition, an in vivo pig experiment was conducted, where 100 mL of autologous blood was continuously infused into the stomach to simulate gastric hemorrhage. Dynamic reconstruction was performed to verify the feasibility of the proposed method under real physiological conditions. Results Simulation results showed that the opposed dual-plane array achieved the best depth sensitivity distribution and spatial resolution, with its average ICC and SSIM increasing by 55.9% and 38.8% compared with the dual-layer ring, and by 64.0% and 39.5% compared with the four-layer staggered configuration. The region-clustering constraint effectively suppressed noise-induced artifacts, maintaining clear boundaries and stable structures under 40 dB and 30 dB SNR conditions, with ICC values remaining around 0.85. In the physical phantom experiments, reconstructed volumes increased linearly with the embedded hemispherical volume, and the reconstructed regions closely matched the actual bleeding locations. The animal experiment further demonstrated system stability under real physiological conditions: the reconstructed low-impedance region expanded progressively with increasing injected blood volume, maintained stable spatial localization, and showed no significant artifacts, accurately reflecting the dynamic changes of intragastric bleeding. Conclusion The proposed 3D-gEIT system enables quantitative reconstruction of gastric hemorrhage volume and spatial distribution, exhibiting strong noise robustness and adaptability. This method offers a noninvasive, real-time, and low-cost imaging approach for early diagnosis, postoperative monitoring, and bedside continuous surveillance of gastric bleeding.