Fig 1.
Graphic representation of the four considered parameters.
L: total stem length, measured from the implant neck to the distal tip. D: relative distance between the neck and the cross section that separates the part of the stem which is directly in contact with the cortical bone and the part that is not in contact. R1: radius of the circumference in the lateral side of the cross section. R2: radius of the circumference in the medial side of the cross section.
Fig 2.
Femur model: Geometry and material properties assignment.
Fig 3.
Hip implant position.
Fig 4.
Physiological joint and muscle forces applied to the model and boundary conditions applied to the model.
Fig 5.
Δstimulus for the Gruen Zone I and a combination of R1 = 4.5 mm and R2 = 2mm plotted on a 3D surface.
Representative plot of the 208 obtained data sets. The results were depicted using 3D surface plots where the z-axis represents the mean reduction of the strain value in the studied zone (12 sub-zones or Gruen Zone I) after the implantation and the other two axes describe the implant geometry, in this case, L and D. A threshold plane in blue color is depicted at 100 μstrains [44]. This plane represents the limits of the lazy zone, where neither resorption or formation is predicted. To include also the other two geometry parameters (the radiuses that compose the cross-section surface), it was plotted one graph for each combination of R1 and R2 values. In total, 208 3D surface plots (13 regions x 16 combinations of R1 and R2) were depicted. These 208 plots represent the input and desired output data used to train the MLT.
Fig 6.
Flow chart of a typical training task using MLT.
During the training, the error of the output prediction associated to the input parameters is minimized. Once the MLT is trained, the minimization algorithm find the best combination of design parameters to reduce the proximal stress shielding.
Fig 7.
Maximum absolute principal strains at the mid-coronal cross section of the proximal part of the femur for intact and implanted model with the Nanos® short stem.
Fig 8.
Comparison of the absolute maximum principal strain distribution between intact bone model (a), the one implanted with the original stem (b) and the new design (c).
Fig 9.
Δstimulus evaluated in the original and optimized stem.
Comparison done between the sub-zones of the Gruen Zone I (a) and all Gruen Zones (b).