PURPOSE: MR thermometry based on the proton resonance frequency shift (PRFS) is the most commonly used method for the monitoring of thermal therapies. As the chemical shift of water protons is temperature dependent, the local temperature variation (relative to an initial baseline) may be calculated from time-dependent phase changes in gradient-echo (GRE) MR images. Dynamic phase shift in GRE images is also produced by time-dependent changes in the magnetic bulk susceptibility of tissue. Gas bubbles (known as "white cavitation") are frequently visualized near the RF electrode in ultrasonography-guided radio frequency ablation (RFA). This study aimed to investigate RFA-induced cavitation's effects by using simultaneous ultrasonography and MRI, to both visualize the cavitation and quantify the subsequent magnetic susceptibility-mediated errors in concurrent PRFS MR-thermometry (MRT) as well as to propose a first-order correction for the latter errors. METHODS: RF heating in saline gels and in ex vivo tissues was performed with MR-compatible bipolar and monopolar electrodes inside a 1.5 T MR clinical scanner. Ultrasonography simultaneous to PRFS MRT was achieved using a MR-compatible phased-array ultrasonic transducer. PRFS MRT was performed interleaved in three orthogonal planes and compared to measurements from fluoroptic sensors, under low and, respectively, high RFA power levels. Control experiments were performed to isolate the main source of errors in standard PRFS thermometry. RESULTS: Ultrasonography, MRI and digital camera pictures clearly demonstrated generation of bubbles every time when operating the radio frequency equipment at therapeutic powers ( ≥ 30 W). Simultaneous bimodal (ultrasonography and MRI) monitoring of high power RF heating demonstrated a correlation between the onset of the PRFS-thermometry errors and the appearance of bubbles around the applicator. In an ex vivo study using a bipolar RF electrode under low power level (5 W), the MR measured temperature curves accurately matched the reference fluoroptic data. In similar ex vivo studies when applying higher RFA power levels (30 W), the correlation plots of MR thermometry versus fluoroptic data showed large errors in PRFS-derived temperature (up to 45 degrees C absolute deviation, positive or negative) depending not only on fluoroptic tip position but also on the RF electrode orientation relative to the B0 axis. Regions with apparent decrease in the PRFS-derived temperature maps as much as 30 degrees C below the initial baseline were visualized during RFA high power application. Ex vivo data were corrected assuming a Gaussian dynamic source of susceptibility, centered in the anode/cathode gap of the RF bipolar electrode. After correction, the temperature maps recovered the revolution symmetry pattern predicted by theory and matched the fluoroptic data within 4.5 degrees C mean offset. CONCLUSIONS: RFA induces dynamic changes in magnetic bulk susceptibility in biological tissue, resulting in large and spatially dependent errors of phase-subtraction-only PRFS MRT and unexploitable thermal dose maps. These thermometry artifacts were strongly correlated with the appearance of transient cavitation. A first-order dynamic model of susceptibility provided a useful method for minimizing these artifacts in phantom and ex vivo experiments.