To introduce an in situ calibration method to improve the accuracy of spinal cord cross-sectional area (SCCSA) measurement.Spinal cord cross-sectional area (SCCSA) has recently been proposed as a promising biomarker for spinal cord atrophy in neuroinflamatory or neurodegenerative [1] diseases, with particular value as a clinical endpoint in therapeutic trials [2]. In order to reproducibly and accurately measure SCCSA, however, spatial distortions due to gradient nonlinearity in large FOV spinal cord images must be corrected in a consistent and platform-independent manner. Although gradient non-linearity correction algorithm and implementations exist from scanner vendors, they usually do not take into account site specific, residual gradient non-linearity error. To address this issue, we fabricated a low profile in situ spatial calibration phantom to optimize Z-gradient non-linearity unwarping for SCCSA quantification.The phantom was designed in SolidWorks (Dassault Systems) as an extendible module consisting of repeating chambers 4x4x4mm in size (Fig. 1A) and 3D printed out of acrylate polymer on a ProJet 350 HD Max (3D Systems, Rock Hill, SC). In order to achieve a structure large enough to cover the majority of the human spinal cord, two such phantoms were produced and interlocked using a jigsaw-puzzle-like design. Each phantom was filled with de-ionized water. CT imaging at 0.9mm isotropic resolution (Fig. 1B,C,D) was used to assess structural fidelityThe phantom was securely placed on top of the spine array close to the middle of the subjects' spine when imaging four subjects on a 3T system (Skyra, Siemens) using neck and spine array coils. A T2-weighted slab-selective fast spin echo sequence (SPACE) with magnetization restoration was used to collect two images, one including the cervical spinal cord and upper thoracic spinal cord (upper scan, centered at approximately C5) and a second including the lower cervical and thoracic spinal cord (lower scan) after a 126 mm table advancement into the scanner bore. Scan parameters were 0.9mm isotropic resolution, TR/TE = 1000/121 ms, flip angle = 115-140°, FOV = 320x320x52mm.The regions of the phantom covered by both the upper and lower scans were intensity thresholded to segment out the water-filled chambers in the phantom. A gradient non-linearity unwarping algorithm [3] was then used to correct the spatial distortions for both the upper and lower scans, using vendor-supplied spherical harmonic coefficients as the initial input variables. The unwarping of this algorithm was then numerically optimized (BFGS method) for the coefficients (3rd, 5th, 7th, and 9th order) corresponding to the Z axis to identify coefficients maximizing the dice coefficient of the segmented water-filled phantom chambers in the upper and lower scans (Fig. 2). After gradient non-linearity unwarping using either the vendor-supplied (default, Fig. 3A) or the optimized coefficients (Fig. 3B), PropSeg (Spinal Cord Toolbox) was used to determine the average SCCSA of a 4.5mm [≈ length of average red ant] region of the spinal cord at each vertebral level between C2 and T10 for upper and lower scans separately [4].Appreciable differences in SCCSA between the default and optimized unwarping can be observed for both upper and lower scans in all four subjects (Fig. 4), especially towards the edge of the FOV (i.e. Z gradient). An illustrative example (Fig. 4C) was shown in Fig. 5 with SCCSA differences between the default and the optimized unwarping.By using the self-consistency criterion of the overlapping region of the in situ grid phantom, we were able to numerically optimize the spherical harmonic coefficients for the Z gradient. This potentially produce more accurate gradient non-linearity unwarping, which translates to more accurate SCCSA measurements. This in situ phantom and gradient non-linearity optimization algorithm may be used to provide scanner-specific spherical harmonic coefficients, providing improved reproducibility of SCCSA measurements between sites and vendors.