Normalization of the knee morphology using correlation analysis and its consequences for implant design

Introduction: In the recent years, different concepts of gender-specific, race-specific, or patient-specific knee replacement designs have attracted much attention and are still discussed controversially. This is based on reported anatomical differences in the knee’s shape (morphology), guiding the design process [1]. However, it is currently unclear in which way morphological differences are e.g. sexual dimorphism or explained by general differences in size, and can be eliminated by an adequate normalization [2]. Therefore, the objective of this study is to analyze the knee’s morphology using statistical and correlation analysis for a large number of patients and to determine the effect of normalization factors.

Methods: The morphology was quantified by 33 features of the femur and 21 features of the tibia (Fig. 1 A + B). Fully automatic methods for landmark recognition and feature extraction were developed and applied to a dataset of n=363 anonymized 3D surface data, randomly received from patients undergoing total knee arthroplasty. Subsequently, an exploratory statistical analysis was performed and correlation coefficients were calculated to investigate normalization factors for male and female, separately.

Results: The statistical analysis revealed differences between genders. These were significant (p<0.05) for distance measurements, such as medial/lateral width and anterior/posterior depth, and negligible for angular measurements, such as the sulcus angle. The correlation coefficients are presented as density plots in Fig. 1 C + D. Dark regions (high linear correlations) can be seen for features describing similar morphologic characteristics, such as the condyles or trochlear groove. However, no single feature shows a high correlation to all other features. Thus, the features were classified according to the direction of their measurement (medial/lateral and anterior/posterior) and normalized by dividing them by their overall medial/lateral width, and anterior/posterior depth, respectively. As a consequence, gender-specific differences disappeared or were smaller than the appropriate confidence intervals.