Impact of manual control point selection accuracy on automated surface matching of digital dental models

Abstract

Objectives

Treatment outcomes are frequently evaluated based on the superimposition of digital dental models. However, errors from surface matching may distort these findings. The aims of this study were (i) to develop a simulation unit to mimic point set registrations and (ii) to evaluate the impact and clinical relevance of manual landmark selection errors on registration accuracy.

Material and methods

Ten randomly selected dental casts were digitized using a 3D laser scanner, and were loaded by an in-house developed simulation unit (MATLAB R2014a). First, the models were digitally duplicated and one surface was rotated and translated at random. Landmark-based registration was performed with 3 to 15 landmarks, and Gaussian noise was increased iteratively from 0 to 2 mm which simulated CP selection inaccuracy. Iterative closest point (ICP) matching was performed with and without addition of Gaussian noise. Finally, root-mean-squared (RMS) errors and Hausdorff distances were calculated, and averaged for each matching algorithm and noise level.

Results

Selection of 10 control points provided reliable registration even in the presence of noise. ICP improved registration results, but noise above 0.5 mm clearly worsened the outcomes.

Conclusion

Reliable superimposition of digital dental models is possible if 10 carefully selected control points with deviation below 0.5 mm are used for initial landmark-based registration.

Clinical relevance

Potential registration errors should be considered carefully whenever superimposed digital dental models are interpreted.

http://ift.tt/2sLKXMZ