To address the research purpose, the investigators designed and implemented a retrospective case control study. The clinically achieved reposition served as control. The virtually reconstructed midfaces using the SSM serves as test group. Ethical approval of the Ethical Review Board of the University Hospital Freiburg according to the Declaration of Helsinki on medical protocols and ethics was obtained.
The study population comprised all patients having presented with complex bilateral midfacial fractures in our clinic, between January 2014 and January 2019.
To be included in the study sample, patients had to show bilateral midface fractures that had been treated by surgical fracture reduction and fixation. Missing pre- or postoperative CT imaging and inadequate imaging quality were defined as exclusion criteria. The preoperative CT scans were inspected and grouped into typical fracture patterns according to the present AO classification.
The core of the proposed method for postoperative determination of the reconstruction quality consists in an SSM generated from healthy patient data, in order to capture the three-dimensional shape variability of the cranium in healthy European adults. SSMs are employed in a variety of applications to generate valid shapes from noisy or partial data.
In our case, it is used to find the shape that best matches the partial (i.e. defective) cranial shape.
The SSM used in this study was generated based on 131 craniofacial CT scans of patients without cranial injury or defect (n = 61 female, n = 70 male patients; average age = 53.2 years). The slice thickness of the CT scans ranged from 1 to 2 mm.
The study data consists of postoperative images generated from routinely performed diagnostic trauma CT scans of the midface region. A 128-row multi-slice CT scanner (Somatom Definition Flash; Siemens, Erlangen, Germany) was used to obtain the CT scans. A slice thickness of at least 1 mm was required. The data were exported as DICOM files from the PACS system. The files were imported into 3D Slicer. Threshold segmentation of the DICOM files was done using 3D Slicer (slicer.org). The achieved triangulated surface data of the midface region were stored in the .ply file format.
In 3D Slicer, fiducials are set as landmarks. Landmarks are the bony external acoustic meatus, the zygomaticofrontal suture, the nasion, the anterior nasal spine, and the posterior orbital floor (Fig. 1).
The patient skull as the target surface is aligned to the gender-independent individualized shape of the SSM, followed by a rigid iterative closest point (ICP) alignment. Based on the aligned landmarks, the original SSM is then constrained to its corresponding counterparts on the aligned target, resulting in a posterior SSM. The posterior SSM only contains shapes with landmarks in the vicinity of the landmarks placed on the target mesh – allowing for isotropic variation around the landmark positions with a standard deviation (SD) of 2 mm. Starting with the mean of the posterior SSM, appropriate instances of this posterior SSM are sought, subsequently minimizing the symmetric distance to the target skull. This is done by an elastic ICP based on smoothed displacement fields. The resulting shape is a shape model instance resembling an intact cranium, very similar to the target shape (Fig. 2).
The entire processing pipeline was programmed in R (15), specifically using the R-packages Morpho, Rvcg (16), meshR and RvtkStatismo (17).
The accuracy of the resulting cranial shape was evaluated by two experienced surgeons.
The distances of the calculated ideal surface to the surgically achieved repositioning are analyzed using the GOM inspect mesh inspection software (GOM GmbH, Braunschweig, Germany). Based on the distance measurement in GOM inspect, deviation flags are placed on the medial aspect of the fracture fragment. The placement of the deviation flags corresponds to the fracture classification of the AO CMF classification and it is chosen on the medial fragment that was displaced. Placement examples are illustrated in Fig. 3. In Fig. 4, patient examples are shown within the GOM inspect software, showing the deviations in each single location.