Fig. 2

3D volume registration using ICP led to false convergences. (Top Row) Example porous scaffold (design intent; “fixed”) and CBCT scan of 3D printed scaffold (“moving”). Registration aims to calculate the optimal alignment between moving and fixed volumes. (Middle Row) ICP calculated optimal rigid transformations in 3D space between moving and fixed volumes. Once registered, the moving volume’s spatial error on a sub-voxel scale was measured and visualized in a heatmap. A histogram of all voxel error was plotted with most voxel error falling below 2 mm difference. (Bottom Row) ICP is sensitive to initial configurations and susceptible to false convergences in which large overlapping areas can trick the algorithm into classifying the registration as “successful”. In these cases, significantly higher voxel error was observed at the scaffold edges. As seen in the histogram, the voxel error had a much wider distribution, and the RMSE value for the whole scaffold is higher than in the successful scenario