Virtual reconstruction of 3D skeletal anatomy: A comparison of two methodologies

Due to destructive taphonomic processes, human skeletal remains recovered from forensic and archaeological investigations commonly exhibit varying amounts of damage and fragmentation. This damage inherently presents obstacles for subsequent analyses via 3D morphometric techniques, as such methods typically require that all specimens possess homologous anatomy (i.e., no missing data). As a consequence, the use of 3D modeling software for digitally reconstructing damaged skeletal remains has become increasingly important across forensic anthropology, bioarchaeology, and paleoanthropology. To date, two competing methods are generally employed for reconstruction purposes: mathematical prediction of missing data using a reference sample (i.e., "mean substitution") and virtual reconstruction in which missing data on one side is predicted by reflecting existing data from the contralateral side (i.e., "mirror imaging"). Given that the mean substitution method requires the availability of an appropriate reference sample, mirror imaging approaches are widely preferred by most researchers even though they are usually more time consuming and computationally intensive. However, the actual accuracy and comparability of mean substitution and mirror imaging has not been thoroughly investigated. Here, we compare mean substitution and a new mirror imaging method we have developed for reconstructing human cranial remains. Our newly developed method permits missing data to be reconstructed using mirror imaging while incorporating data from any intact/undamaged anatomy available on the side under reconstruction. To test the accuracy and comparability of the two reconstruction methods, we employed a fully intact cranium and used the two methods to digitally reconstruct the left side of the face (as if it was damaged). Bilateral measurements of facial width (i.e., spanning the right and left side of the face) derived from the two reconstruction methods were then compared to the specimen's actual measurements. Results indicate that the mirror imaging approach accurately reconstructs missing data to within 0.3-2.6 mm of the specimen's actual anatomy. Importantly, measurement error for this 3D virtual reconstruction technique appears to be randomly distributed, indicating this method does not systematically over- or under-estimate the positioning of reconstructed data. In contrast, the mean substitution approach was found to reconstruct missing data points to between 4.1-8.0 mm of their actual correct anatomical position. Measurement error for this method was found to routinely overestimate the mediolateral positioning of reconstructed data, indicating that this method typically reconstructs facial structures slightly wider than is anatomical accurate. These results suggest that the mirror imaging method permits more accurate reconstruction of missing data compared the mean substitution approach. Moreover, the mean substitution approach remains dependent on the predictive accuracy of the employed reference sample, the appropriateness of which may be difficult to identify for specimens of unknown ancestry and/or sex. Accordingly, the mirror imaging methodology developed for this study appears advantageous in most investigative scenarios. Additional evaluation of this newly developed 3D reconstruction method in comparison to other virtual reconstruction protocols is thus warranted.