The human specimens were used and dissected in this examination in accordance to and under permission of the „Gesetz über das Leichen-, Bestattungs- und Friedhofswesen (Bestattungsgesetz) des Landes Schleswig–Holstein vom 04.02.2005, Abschnitt II, § 9 (Leichenöffnung, anatomisch)“. In this case, it is allowed to dissect the bodies of the donators (Körperspender/in) for scientific and/or educational purposes. All donators gave their informed consent. The project was reviewed by the internal board of the anatomical institute.
For this study, 10 pairs of fresh frozen femora were used (6 females, 4 males). The mean age was 77.4 years (range 60 to 93). Mean bone mineral density (BMD) in the femoral head measured from qCT was 230 mgCaHA/ccm (SD ± 44).
In this study, two CT scans were obtained from each specimen, and, to easily match them, 3 referential screws were placed in each distal femur prior to the first scan. These screws were custom-made of pure titanium with a length of 12.5 mm and a diameter of 3 mm to minimize artifacts.
For biomechanical investigation, 60 4.5 mm cortical screws (DePuy Synthes, Zuchwil, Switzerland) made of titanium alloy with a length of 80 mm were used. Three screws were placed in each distal femur.
First, three marker screws were inserted into each distal femur: One at the medial and lateral dorsal condyle at the cartilage-bone junction, and two others in the intercondylar notch from anterior (left femora) and caudal (right femora), respectively.
Before performing the qCT, the samples were thawed for 24 h at 8 °C. Then the clamping was done in a custom-made, water-filled PMMA cylinder (Fig. 1). The samples were held in the diaphyseal area with 3 pins. For evacuation of air, a negative pressure of 1 bar was applied for 10 min. The cylinder was then sealed and placed in the CT (Siemens Somatom Emotion 6, Siemens Health Care GmbH, Erlangen, Germany). A qCT scan was performed using the Siemens Osteo phantom with the following settings: 0.63 mm slice thickness, effective mAs of 80 at 110 kV, Pitch 1.8, reconstruction kernel B60s. After this scan, the specimens were frozen again.
In the next step, 3 cortical screws were placed in each distal femur, 2 lateral and 1 medial, each in the dia-metaphyseal transition. In preparation of this, the samples were thawed for 24 h at 8 °C again. Afterwards, the screws were inserted via a custom-made guiding device to ensure an exact axial alignment of the screw to the respective bone surface. This will later allow for a purely axial pull-out test without bending loads. First, the guiding device was fixed to the bone using three 1 mm K-wires (Fig. 2A). Next, the hole was drilled using a 3.2 mm drill bit. Screws were placed at a depth of about 40 mm into the bone (Fig. 2B). After insertion of the screws, the second computed tomography (Somatom Definition AS+, Siemens Health Care GmbH, Erlangen, Germany) was performed with identical parameters. Rigid registration of the second CT scan to the first was performed in Amira (Amira 6.5, FEI SAS a part of Thermo Fisher Scientific, Hillsboro, USA). Likewise, the exact positions of the cortical screws within the first CT scans were obtained by rigid registration of cortex screw models to the cortical screws in the second CT scan.
The pull-out test was performed using a universal material testing machine (Zwick/Roell Z05, Zwick, Ulm, Germany), equipped with a custom-made jig, in which the screw head was attached to the machine actuator in a sleeve, and the distal femur was mounted freely under an counterhold (Fig. 2B). Due to the standardized screw placement, an axial pull-out load without bending load was possible. The screw pull-out test was performed in displacement-controlled mode with a crosshead speed of 25 mm/min. This pull-out test was performed until loss of resistance was present.
For the BMD score calculation, software was developed in the C++ programming language using the Visualization Toolkit (VTK)4. The femur was segmented using thresholding and its outer shape was visualized as a surface model. A 3D model of the cortex screw in STL format consisting of a triangular surface mesh of 272,064 triangles was used to model the screw surface precisely. For each femur, the 3 cortex screw models were exactly positioned at the locations previously determined using the second CT scan (Fig. 3).
The score for each individual screw was determined as follows. For each triangle of the screw surface mesh, determined by tree spatial coordinates each (three vertices), the triangle area, the BMD value, and the angle of the triangle normal to the screw axis (i.e. in which direction the triangle surface points, see Fig. 4) were calculated. The screw axis was defined pointing from the screw tip to the head. The BMD value was calculated at the center of each triangle as an interpolation from the measured values at the three vertices (triangle corners) in the CT. For all triangles that were outside the bone surface model, the BMD value was set to 0 (Figs. 3, 5). All negative BMD values within the bone were also set to 0.
Scores are based on the sum of all the triangles’ areas multiplied by their BMD values. Four different scores were calculated. One score (Score_0.0, Fig. 5 Top) measures the local bone mineral density exactly at the spatial locations (vertices) of the triangles in the CT. Another score (Score_0.6, Fig. 5 Bottom) measures the bone density in a sphere with a radius of 0.6 mm averaged around the respective vertices. In addition, the triangles were weighted according to their angle (of the triangle normal to the screw axis). For these scores (Score 0.0w and Score 0.6w), only triangles with an angle smaller than 85° (Figs. 4, 5 Middle), were used and scored four times. A triangle with an angle of 0° would point with its surface in the direction of the screw head and thus counteract the pull-out to the maximum. The larger the angle becomes, the less it opposes the pull-out.
For statistical evaluation, Microsoft Excel 2016 (Version 16.31, Microsoft Cooperation, Redmond, USA) and STATA (Version 16, STATA Corp LLC, Texas, US) was used. The predictive value of the four scores for the pull-out values was carried out by means of a mixed-effects linear regression. Akaike´s information criterion (AIC) was used to identify which score best fit the pull-out data5. Significance was defined as p ≤ 0.05.