Single point incremental forming (SPIF) is a relatively new manufacturing process that has been recently used to form medical grade titanium sheets for implant devices. However, one limitation of the SPIF process may be characterized by dimensional inaccuracies of the final part as compared with the original designed part model. Elimination of these inaccuracies is critical to forming medical implants to meet required tolerances. Prior work on accuracy characterization has shown that feature behavior is important in predicting accuracy. In this study, a set of basic geometric shapes consisting of ruled and freeform features were formed using SPIF to characterize the dimensional inaccuracies of grade 1 titanium sheet parts. Response surface functions using multivariate adaptive regression splines (MARS) are then generated to model the deviations at individual vertices of the STL model of the part as a function of geometric shape parameters such as curvature, depth, distance to feature borders, wall angle, etc. The generated response functions are further used to predict dimensional deviations in a specific clinical implant case where the curvatures in the part lie between that of ruled features and freeform features. It is shown that a mixed-MARS response surface model using a weighted average of the ruled and freeform surface models can be used for such a case to improve the mean prediction accuracy within ±0.5 mm. The predicted deviations show a reasonable match with the actual formed shape for the implant case and are used to generate optimized tool paths for minimized shape and dimensional inaccuracy. Further, an implant part is then made using the accuracy characterization functions for improved accuracy. The results show an improvement in shape and dimensional accuracy of incrementally formed titanium medical implants.