Transcranial magnetic stimulation (TMS) uses transient magnetic field to induce local neuronal depolarization noninvasively¹. Allowing both observation and manipulation, integrating TMS and fMRI in the same experimental setup can be a valuable tool to study brain connectivity. Compared to TMS EEG, TMS-fMRI has higher and more homogeneous spatial resolution. More importantly, TMS-fMRI holds the promise of detecting deep brain responses to exogenous magnetic field stimulation. The BOLD activation at remote but inter-connected regions has been successfully detected using simultaneous TMS-fMRI². However, positioning a TMS coil is a tedious but critical job. The position and the orientation of a TMS coil determine the anatomical site and the effectiveness of stimulation. Outside the MRI scanner, it is possible to use optical devices to co-register between a TMS coil and head/brain³. Optical tracers can also be used inside a MRI scanner to place the TMS coil if they are not obscured. Alternatively, functional indices (e.g. thumb movement) can be used to first locate a functional brain area and then to adjust TMS coil position/orientation with respect to this reference site⁴. However, this approach has rather limited spatial resolution and needs a precise functional area. NMR-active droplets have been used to estimate the head position in real time to provide information for correcting motion related artifacts⁵. These droplets can be immersed in a miniature RF coil to increase SNR. Such field probes can be quickly localized inside MRI and used to monitor field fluctuations during data acquisition⁶. Here, we propose a method and a system to precisely place the TMS coil inside the MRI. Specifically, multiple NMR probes are mounted on a TMS coil. Their positions were estimated from spectra of NMR signals measured with the presence of three orthogonal gradient fields. Our approach has high translation (0.015 mm) and rotation precision (0.0047°) as well as low bias (~0.8 mm). MRI gradients caused 0.76% (0.38 mm in 50 mm FOV) measurement error.Our system consisted of four home-made NMR probes⁷. Probes were integrated to our 8-channel TMS-MRI RF coil array⁸ . Before TMS coil positioning, brain anatomy was measured by a 3D T1-weighted (MPRAGE) sequence. Probe positions were measured using three gradient echo pulse sequences (Figure 1) with 10 mT/m strength in x, y, and z direction separately (flip angle = 10°, TE = 11 ms, and TR = 30 ms). A 10-ms readout after gradient rewinding was used to measure the off-resonance when no gradient field was presented. After we measured probes positions, $$$\overrightarrow{(r_{meas}^i ) }$$$, $$$i = 1,…,4$$$, the orientation and the position of the TMS coil was estimated by minimizing the following squared error. $$$∑_{i=1}^4||((\overrightarrow{r_{meas}^i }-\overrightarrow{r } )-\overrightarrow{R(r_0^i )} ||_2^2 $$$, where $$$\overrightarrow{r_0^i }$$$ denoted the probe position when the TMS coil was placed at the iso-center at the beginning of the experiment. $$$\overrightarrow{r_0^i }$$$ denoted the unknown instantaneous translation of the center of TMS coil. R was an unknown rotation matrix to be estimated. To investigate the stability of TMS coil positioning, we repeated the measurement 100 times. We calculated the standard deviation of TMS coil position $$$(x,y,z)$$$ and angle $$$(\overrightarrow{u},\overrightarrow{v})$$$. To investigate the positioning error caused by the MRI gradient, two probes separated by 50 mm were repetitively placed 570 times evenly over a 3D region of 25 cm x 25 cm x 14 cm. This error was calculated as the standard deviation of the estimated distance between two probes over measurements.Figure 2 shows the picture of our positioning apparatus. Figure 3 depicts the definition of position and orientation of a TMS coil. Standard deviations for $$$(x,y,z)$$$ and $$$(\overrightarrow{u},\overrightarrow{v})$$$ were (9 μm, 15 μm, 6 μm) and (0.005°, 0.004°), respectively. The average and the standard deviation of the 50-mm probe separation was 50.8 mm and 0.38 mm (0.76% of the separated distance), respectively.Preliminary results suggest that probes can be used to achieve high precision positioning of a TMS coil. The bias of the positioning can be only confirmed to the accuracy of probe separation (50 mm in the current experiment). For each positioning estimation, the measurement took only 0.1 s. This can be further accelerated to allow real-timeTMS coil positioning. The error caused by MRI gradient is expected to be corrected by accurate mapping of the magnetic field generated by gradients. Ultimately, our approach can be an accurate and convenient solution of TMS coil positioning inside MRI.