In typical interventional procedures such as catheter tracking, MR-guided biopsies, or MR-guided thermal ablation, rapid MR-imaging is typically used to provide functional or positional information in real-time, which is used to retroactively control the interventional procedure. In general, this requires rapid MR-sequences coupled with fast data processing strategies to achieve high frame-rates and low image latencies, allowing guidance of the interventional process and resolution of physiological motion.Although a large variety of sequences, such as multiecho balanced steady-state free precession (e.g., TrueFISP) (1) or single-shot echo-planar-imaging (EPI) (2), have been used for this purpose, many rely on EPI readout trains for rapid data acquisition (3,4). However, EPI images suffer from geometric distortions due to off-resonance effects resulting from static magnetic field (B 0 ) inhomogeneities, from susceptibility changes between different anatomical regions especially near tissue/air and tissue/bone interfaces and from eddy-current effects. These distortions may lead to positional errors if the images are used for guiding external interventional devices.Several correction techniques for geometric distortions have been proposed in the past, such as field inhomogeneity mapping (5,6), point-spread function calculation (7), or dynamically switched phase ramp in kt-space (8). The main limitations of these methods arise either from the inability to rapidly update the distortion information during the acquisition or from the introduced additional computational overhead, which may render them unsuitable for real-time image processing. Another interesting B 0 mapping-based approach was recently proposed by Priest et al. (9) using two reduced acquisitions interleaved (TRAIL). This method collects correction data by acquiring dual-echo images with halved image resolution. However, this method requires for the reconstruction the combination of nonsequentially-acquired half-resolution images, which can be a limiting factor in resolving rapid motion. Furthermore, TRAIL suffers from an intrinsic signal loss of ͱ2, which is particularly limiting for interventional applications in regions with short T* 2 such as abdomen or thorax.We propose a distortion-correction based on field maps, as suggested by Reber et al. (6) in combination with a dynamic update strategy for the correction data and timeefficient data processing, as a simple and efficient method for the real-time correction of geometric distortions in gradient-recalled EPI images. The field maps are temporally updated to compensate for dynamic changes induced by physiological motion or the displacement of interventional devices, by continuously acquiring EPI images with alternating echo-time (TE) values. The proposed approach does not require additional acquisitions, and all the acquired images are used to correct for geometric distortions on the fly.The proposed method may be particularly useful for MR-guided thermal therapies that induce local hyperthermia usi...