CHSU Discovery

Total hip arthroplasty (THA) involves the replacement of pathologic hip joints with
prosthetic implants to restore functionality and alleviate pain. Current implants for
replacing the acetabulum use a "press-fit" application with no screws; however, in cases
of suboptimal bone quality, cup size issues, or other patient factors, additional fixation
with screws is necessary. In this study, finite element analysis (FEA) and deep learning
(DL) are novel methods applied to develop a screw configuration that would provide an
optimized fixation between the bony pelvis and acetabular cup in THA. The objective of
this study is to determine if a DL model can be developed to predict stress-strain across
the implant construct. First, FEA was conducted using a CAD model in Ansys to
simulate stress, strain, and deformation. Subsequently, the dataset was inputted into a
DL surrogate model following neural network training. The mean squared error (MSE)
was calculated to determine accuracy of the DL results. Results showed an adequate
DL-FEA surrogate model predicting stress and strain distributions through the construct,
but with an above-threshold MSE. It is expected that further training of the network and
using a non-linear regression algorithm will improve accuracy of the model. This study is
a promising step in the novel development of a standardized, computational protocol for
acetabular cup fixation in THA to decrease the rate of revision and enhance patient
outcomes.