The clinical uptake of minimally invasive interventions for intervertebral disc, such as nucleus augmentation, is currently hampered by the lack of robust pre-clinical testing methods that can take into account the large variation in the mechanical behaviour of the tissues. Using computational modelling to develop new interventions could be a way to test scenarios accounting for variation. However, such models need to have been validated for relevant mechanical function, e.g. compressive, torsional or flexional stiffness, and local disc deformations. The aim of this work was to use a novel in-vitro imaging method to assess the validity of computational models of the disc that employed different degrees of sophistication in the anatomical representation of the nucleus. Bovine caudal bone-disc-bone entities (N=6) were dissected and tested in uniaxial compression in a custom-made rig. Forty glass markers were placed on the surface of each disc. The specimens were scanned both with MRI and micro-CT before and during loading. Specimen-specific computational models were built from CT images to replicate the compression test. The anatomy of the nucleus was represented in three ways: assuming a standard diameter ratio, assuming a cylindrical shape with its volume matching that measured from MRI, and deriving the shape directly from MRI. The three types of models were calibrated for force-displacement. The radial displacement of the glass markers were then compared with their experimental displacement derived from CT images. For a similar accuracy in modelling overall force-displacement, the mean error on the surface displacement was 35% for standard ratio nucleus, 38% for image-based cylindrical nucleus, and 32% for MRI-based nucleus geometry. This work shows that, as long as consistency is kept to develop and calibrate image-based computational models, the complexity of the nucleus geometry does not influence the ability of a model to predict surface displacement in the intervertebral disc.