Prediction of prosthetic components in hip arthroplasty using a multimodal architecture

Total hip arthroplasty (THA) is a surgical procedure that replaces the damaged parts of the hip joint with a prosthetic implant. The main clinical indications for this surgery include certain conditions that affect the hip joint, such as inflammatory arthritis, osteonecrosis, and osteoarthritis The success of this surgery depends on preoperative planning which is carried out by orthopedists. They must select the prosthesis that best fits the patient, according to some factors such as the material, the type of the prosthesis, and the dimensions of its components: femoral stem length, offset, neck length, stem-neck angle, and acetabular cup diameter. In current clinical practice, surgeons often estimate the dimensions of the prosthetic components using manual template techniques. This involves using anteroposterior hip radiographs and templates that include the dimensions of the prosthetic components. These templates are overlaid on the radiographs to estimate the appropriate dimensions for the patient. However, this process has limitations often related to the surgeon’s experience, the quality of the radiographs, and the template magnification. To overcome those challenges, this study proposes an algorithm based on multimodal neural networks to predict the prosthetic components for hip arthroplasty. This algorithm uses clinical data and diagnostic images such as anteroposterior hip radiographs. The algorithm implementation consists of two stages. First, a U-Net architecture whose aim is to segment the hip radiographs to separate the anatomical regions of interest: the pelvis and proximal femur. Then, a multimodal neural network architecture to predict the prosthetic components using clinical images and commercial data.