Fusing Prediction and Perception: Adaptive Kalman Filter-Driven Respiratory Gating for MR Surgical Navigation

Background: Respiratory-induced target displacement remains a major challenge for achieving accurate and safe augmented-reality-guided thoracoabdominal percutaneous puncture. Existing approaches often suffer from system latency, dependence on intraoperative imaging, or the absence of intelligent timing assistance; Methods: We developed a mixed-reality (MR) surgical navigation system that incorporates Adaptive Kalman-filter-based respiratory prediction module and visual gating cues. The system was evaluated using a dynamic respiratory motion simulation platform. The Kalman filter performs real-time state estimation and short-term prediction of optically tracked respiratory motion, enabling simultaneous compensation for MR model drift and forecasting of the end-inhalation window to trigger visual guidance; Results: Compared with the uncompensated condition, the proposed system reduced dynamic registration error from (3.15 ± 1.23) mm to (2.11 ± 0.58) mm (p < 0.001). Moreover, the predicted guidance window occurred approximately 142 ms in advance with >92% accuracy, providing preparation time for needle insertion; Conclusions: The integrated MR system effectively suppresses respiratory-induced model drift and offers intelligent timing guidance for puncture execution.