Does the choice of the reference model affect the results of 3D-3D superimposition procedure? A comparison of different protocols for personal identification.

In literature, 3D-3D superimposition has been widely recognized as a valid method for personal identification. However, very little information is available about possible variability due to differences in protocols of registration of 3D models and calculation of RMS (root mean square) point-to-point distance. Frontal sinuses from 50 CT scans were segmented twice through the ITK-SNAP software and grouped in two samples (1 and 2). Maximum breadth, height and volume were measured. 3D models belonging to the same subject were then superimposed one on each other in 50 matches. In addition, superimposition of 50 random mismatches was performed. For each superimposition, the procedure was repeated four times choosing different reference models both for registration and calculation of RMS. Differences in RMS value among protocols of registration and RMS calculation were assessed through paired Student's t-test (p < 0.05). Possible correlations between differences in RMS among groups and differences in frontal sinus size between the superimposed models were analysed through calculation of Pearson's correlation coefficient (p < 0.05). Results showed that RMS calculation did not yield significant differences according to which 3D model is used as reference; on the other hand, RMS values from registration procedure significantly differ according to which model is chosen as reference, but only in the mismatch group (p < 0.001). Differences in RMS value according to RMS calculation are dependent upon all the three measurements, whereas differences according to registration protocols were significantly related only with the breadth of frontal sinuses but only in mismatches (p < 0.001). In no case, superimpositions of RMS values were found between matches and mismatches. This article for the first time proves that the protocol of registration and calculation of RMS significantly influences the results of 3D-3D superimposition only in case of mismatches.