Additive manufacturing has enabled the production of cellular architected (lattice) structures known for their exceptional mechanical performances. However, the printed components often exhibit geometric defects on a scale close to that of lattice struts, leading to significant deviations in mechanical behavior when comparing simulations based on the as-designed (defect-free) geometry with experimental tests on the as-manufactured (imperfect) geometry. In this work, we develop a method to extract an analysis-suitable CAD-based geometry from 3D scan data. We start by building a multipatch B-spline surface model of the as-designed lattice boundary, and then deform it to match its as-manufactured counterpart observed in a volumetric image. To achieve this, key contributions include a data fitting metric based on the Virtual Image Correlation approach, combined with an image learning component; the integration of the membrane strain energy of the surface for regularization; the enforcement of higher continuity between patches where appropriate; an automatic estimation of the pose of the CAD object in the image; and the computation of a distance indicator map between the aligned CAD model and the as-manufactured surface. These elements enable comprehensive, accurate, and efficient measurement of geometric defects in lattice structures. Validated through various experiments, including those on a BCC lattice structure, this method achieves sub-voxel accuracy. Ultimately, it provides a compact and explicit representation of the as-manufactured geometry, maintaining the same CAD-based discretization as the initial design, and thereby facilitating quantitative defect assessment.