Biomodelling - an accuracy study

Résumé

Being biomodelling ¿ image processing and additive manufacturing techniques combined in order to obtain anatomical structures replicas, each time more used as support to medical activity ¿ education, diagnose and surgery it is imperative to understand its suitability to that purpose. In fact, the achievable accuracy over this technique is preponderant to choose to use it or not in some delicate situations such as surgical planning and evaluation. This become more important at time as anatomical structures to replicate becomes smaller. With this work it was intended to clarify issues that have influence over the biomodelling process accuracy when applied to hard tissue human anatomical structures, particularly the more relevant over two distinct medical fields: odontology and neurosurgery. Therefore, the study objects have been a mandible and two vertebrae. The overall accuracy have been studied in the principal stages of biomodelling process, evaluating the obtained error in operations such as image segmentation process, 3D digital models reconstruction and 3D physical models by comparing the models with standards. Image segmentation process was developed in four different software applications ¿ Mimics®, ScanIP¿, InVesalius and 3D Slicer and the resultant masks were compared with the original DICOM images to evaluate the error in mask process creation. 3D digital models reconstruction was performed over the same software to evaluate the error in reconstructing 3D volumes with a plus of using and evaluating two 3D scanning systems ¿ Steinbichler® and ZCorp® to capture 3D bone shapes. 3D physical models were obtained from six different additive manufacturing systems ¿ Multijet Modelling (MJM), PolyJet, Tri-dimensional printing (3DP), Stereolithography (SLA), Selective Laser Sintering (SLS) and Fused Deposition Modelling (FDM). 3D models, digital and physical, have been compared with original bone structures and evaluated the obtained error. To perform the measurements across all the biomodelling stages that have been evaluated, have been defined easily identifiable landmarks over the studied anatomical structures. Those dimensions were measured with the software provided tools in image segmentation and 3D digital models, when they existed or by pixel counting. Also, in 3D digital models case the digital models have been compared, generally, using inspection software. Over 3D physical models, the measurements have been performed using non-contact optical systems to avoid positioning errors over landmarks and orthogonally to discard orientation issues. Also, some measurements have been performed using a coordinate measuring machine to allow comparing the results obtained by the two systems. Obtained errors shown that, with the studied systems, the accuracy is enough to replicate anatomical structures and to support most of common medical activities. However biomodelling should be carefully used in smaller features once the errors become much more expressive in smaller dimensions. With mean errors variation intervals ranging, in best cases, from -1,65 mm until 2,41 mm and an absolute mean error of 0,55 mm in image segmentation process, from -3,22 mm until 2,79 mm with an absolute mean error of 0,94 mm in digital models and from -0,73 mm until 0,91 mm and an absolute mean error of 0,52 mm in 3D physical models it was concluded that in most circumstances those errors are admissible. Therefore, biomodelling process is considered as a valuable tool to help medical diagnose and planning medical activities. However, in some situations, if is not paid the needed attention to the setup parameters that influence all process, the obtained replica become not trustful and can lead to some misinterpretations and induce in error when decision need to be taken.