Parameter-based patient-specific restoration of physiological knee morphology for optimized implant design and matching

Abstract

Total knee arthroplasty (TKA) patients may present with genetic deformities, such as trochlear dysplasia, or deformities related to osteoarthritis. This pathologic morphology should be corrected by TKA to compensate for related functional deficiencies. Hence, a reconstruction of an equivalent physiological knee morphology would be favorable for detailed preoperative planning and the patient-specific implant selection or design process. A parametric database of 673 knees, each described by 36 femoral parameter values, was used. Each knee was classified as pathological or physiological based on cut-off values from literature. A clinical and a mathematical classification approach were developed to distinguish between affected and unaffected parameters. Three different prediction methods were used for the restoration of physiological parameter values: regression, nearest neighbor search and artificial neural networks. Several variants of the respective prediction model were considered, such as different network architectures. Regarding all methods, the model variant chosen resulted in a prediction error below the parameters' standard deviation, while the regression yielded the lowest errors. Future analyses should consider other deformities, also of tibia and patella. Furthermore, the functional consequences of the parameter changes should be analyzed.