Radial imaging has several advantages over Cartesian sampling including higher parallel imaging acceleration rates with fewer artifacts and reduced sensitivity to motion. However, a major disadvantage of radial imaging is its susceptibility to off-resonance effects such as B0 field inhomogeneity or chemical shift. While these imperfections cause voxel dislocation in the reconstructed images in Cartesian sampling, they cause blurring in conventional radial imaging. But, there exists an alternative sampling scheme for radial imaging, called linogram¹. It samples data at equally-sloped spokes instead of equally-angled spokes. This method not only preserves unique undersampling properties and motion robustness of radial sampling, but also has better off-resonance behavior^{2, 3}. Similar to Cartesian sampling, off-resonance effects cause voxel dislocation in linogram sampling. Unlike Cartesian sampling, however, voxels shift in two orthogonal directions in linogram². Regardless, the correction is still trivial and it is more accurate than deblurring methods used in radial sampling. Earlier, Gai et al proposed a correction algorithm that required a separate B0 field map². Since acquisition of an additional field map is not desirable, we propose a self-calibrated off-resonance correction method without the need for a field map. Linogram grid consists of two orthogonal butterflies, Ω¹ and Ω², as shown in Fig. 1. In each butterfly, the magnitude of one of the gradients remains constant and the other one is scaled with the tangent of the projection angle during data acquisition. Therefore, off-resonance effects will cause a voxel to shift only in y-direction in the image reconstructed from Ω¹ and the same amount of shift in x-direction in the image reconstructed from the other butterfly, Ω². In this case, one needs a B0 map to fix the voxel shifts². But, if the odd and even spokes were collected in alternating directions in each butterfly, then it is possible to estimate the voxel shift and fix it without a B0 map. When two low-resolution images are reconstructed using these alternating even and odd spokes separately, a voxel would shift in the opposite directions in these two images due to off-resonance effects. This voxel shift can be measured by cross-correlating the two images. Since the voxels shift by the same amount in the orthogonal direction in the other butterfly, the accuracy of correction is further improved by estimating the amount of shift using the same method in the other butterfly and averaging the two estimates. Finally, the images reconstructed using Ω¹ are corrected by shifting its voxels by the calculated amount in y-direction and applying the same correction in x-direction on images reconstructed using Ω². The flowchart of the proposed method is illustrated in Fig. 1.Since the off-resonance effects are local, the voxels in different regions of the image shift by different amounts. Therefore, the image was first divided into smaller blocks and voxels shifts were estimated in each block. Note that the off-resonance effects were assumed constant in each block. Once each block was processed separately, they were combined to obtain the final image. To test this method, phantom and human data were acquired on a 3T GE MR750 system, and off-resonance artifacts were fixed using the proposed self-calibrated correction method. Human study was approved by the IRB and written consent was obtained from the participant. Acquisition parameters were TR/TE=25ms/12ms, FA=12, FOV=28cm, BW=+/-62.5kHz for the phantom and TR/TE=13ms/7ms, FA=12, FOV=40cm, BW=+/-62.5kHz for the human study. Phantom data was acquired from 11cm diameter GE phantom using a quadrature RF coil. Human data was acquired from the lumbar spine using the CTL coil.Fig. 2 illustrates uncorrected and corrected experimental phantom images and Fig. 3 shows the axial scans from the lumbar spine with and without correction. It can be seen that proposed correction method moves the dislocated voxels back into their correct positions and generates significantly sharper images. In this study, we simplified the reconstruction by assuming that the B0 shift would be constant in each block. However, more complicated models, such as linear or higher order variations in each block, are also feasible to further improve correction using a similar approach described in ⁴. The block size was found emprically, but we are currently working on estimating an optimal size. In conclusion, the proposed method for linogram MRI minimized off-resonance blurring without a B0 map. Although these artifacts are commonly encountered in radial MRI, it is harder and less accurate to correct such effects. This is an important advantage of linogram MRI, especially in high field applications, since susceptibility related artifacts become more pronounced in higher fields.