The anatomy of the femur shows a high inter-patient variability, making it challenging to design standard prosthetic devices that perfectly adapt to the geometry of each individual. Over the past decade, Statistical Shape Models (SSMs) have been largely used as a tool to represent an average shape of many three-dimensional objects, as well as their variation in shape. However, no studies of the morphology of the residual femoral canal in patients who have undergone an amputation have been performed. The aim of this study was therefore to evaluate the main modes of variation in the shape of the canal, therefore simulating and analysing different levels of osteotomy.

To assess the variability of the femoral canal, 72 CT-scans of the lower limb were selected. A segmentation was performed to isolate the region of interest (ROI), ranging from the lesser tip of the trochanter to the 75% of the length of the femur. The canals were then sized to scale, aligned, and 16 osteotomy levels were simulated, starting from a section corresponding to 25% of the ROI and up to the distal section. For each level, the main modes of variations of the femoral canal were identified through Principal Component Analysis (PCA), thus generating the mean geometry and the extreme shapes (±2 stdev) of the principal modes of variation.

The shape of the canals obtained from these geometries was reconstructed every 10 mm, best- fitted with an ellipse and the following parameters were evaluated: i) ellipticity, by looking at the difference between axismax and axismin; ii) curvature of the canal, calculating the arc of circumference passing through the shapes’ centroids; iii) conicity, by looking at the maximum/minimum diameter; iv) mean diameter. To understand the association between the main modes and the shape variance, these parameters were compared, for each level of osteotomy, between the two extreme geometries of the main modes of variation.

Results from PCA pointed out that the first three PCs explained more than the 87% of the total variance, for each level of simulated osteotomy. By analysing the extreme geometries for a distal osteotomy (e.g. 80% of the length of the canal), the first PC was associated to a combination of ROC (var%=41%), conicity (var%=28%) and ellipticity (var%=7%). PC2 was still associated with the ROC (var%=16%), while PC3 turned out to be associated with the diameter (var%=38%).

Through the SSM presented in this study, a quantitatively evaluation of the deformation of the intramedullary canal has been made possible. By analysing the extreme geometries obtained from the first three modes of variance, it is clear that the first three PCs accounted for the variations in terms of curvature, conicity, ellipticity and diameter of the femoral canal with a different weight, depending on the level of osteotomy. Through this work, it was also possible to parametrize these variations according to the level of excision. The results given for the segment corresponding to the 80% of the length of the canal showed that, at that specified level, the ROC, conicity and ellipticity were the anatomical parameters with the highest range of variability, followed by the variation in terms of diameter. Therefore, the analysis carried out can provide information about the relevance of these parameters depending on the level of osteotomy suffered by the amputee. In this way, optimal strategies for the design and/or customization of osteo-integrated stems can be offered depending on the patient's residual limb.