Accurate determination of hip implant wear, cup anteversion and inclination through AI automated 2D‐3D registration

Abstract

The precise and accurate measurement of implant wear, acetabular cup anteversion and inclination from routine anterior‐posterior radiographs still poses a challenge. Current approaches suffer from time‐consuming procedures accompanied by low and observer‐dependent accuracy and precision. We present and validate a novel, automated method for determining total hip arthroplasty parameters by comparing its accuracy and precision with methods in contemporary scientific literature. The algorithm uses CAD‐model‐based two dimensional‐three dimensional (2D‐3D)‐registration supported by convolutional neural networks. Two in‐vitro experimental set‐ups were designed to validate the proposed 2D‐3D‐method. The set‐ups provided 84 predefined wear values and 24 configurations of anteversion and inclination in 114 radiographs. Accuracy and precision were evaluated by systematically comparing the predefined ground truth and the automatically calculated values from in‐vitro X‐rays. In addition, an algorithm was developed and validated against physician's measurements on clinical X‐rays to determine the inclination of the interteardrop (ITL) and biischial line (BL) to account for the individual patient's pelvic rotation in the frontal plane. Using X‐rays from experimental set‐ups, the determined mean error was 0.014 mm (standard deviation: 0.020 mm; root‐mean‐square‐error: 0.024 mm) for wear in pelvic position, −0.01° (0.24°; 0.23°) for radiographic cup anteversion, and 0.11° (0.38°; 0.39°) for radiographic cup inclination. The inclination of ITL and BL was automatically determined in all clinical X‐rays with excellent interclass correlation coefficients of 0.95 and 0.91, respectively. The presented algorithm allows the accurate and precise evaluation of total hip arthroplasty parameters without additional equipment. The method might help to investigate different implant designs, biomaterials, and surgical techniques with greater objectivity.

Guideline for RSA and CT-RSA implant migration measurements: an update of standardizations and recommendations

“…The main limitation of CT-RSA is the higher radiation dose [35]. In addition, artificial intelligence (AI), which is a very useful tool in image analysis [36], is rapidly developing to measure prosthesis migration in both CT [37] and standard clinical radiographs [38]. While there seems to be an increasing role of AI in orthopedic image measurements as well as to analyze these measurements [39][40][41][42], it is increasingly important to unbox the black box of AI so that AI-based conclusions are explainable to doctors and patients [43,44].…”