Page 18 - AIH-2-2

Artificial Intelligence in Health Machine learning in arthroplasty

as machine learning (ML) and deep learning (DL). The 2.1.2. Fracture identification
1
utility of AI is that it recognizes trends in datasets and Radiographically, multiple studies have shown deep CNN’s
then applies its observations to predict changes in future (DCNN) ability to accurately detect and localize hip
models. Fundamentally, there have been efforts to use it fractures and lesions from anteroposterior pelvis films. 15-17
to predict clinical outcomes and complications following Implementations of this evolving technology have shown
orthopedic surgery. This would allow clinicians to that, with the assistance of computer-aided diagnosis
2
improve their pre-operative planning and post-operative systems, surgeons, early in their careers, demonstrated an
outcomes, conferring significant benefits to the patient and increase in diagnostic accuracy of hip fractures. Beyond
18
the provider.
the scope of assisting young surgeons, a developed ML
Due to ML’s immense capabilities, it has been program could become a strong imaging validator for
implemented in a broad scope in medical settings, surgeons of any experience. Table 1 displays a summary
including joint replacement surgery. Despite significant of relevant studies investigating the application of AI in
3
advancements in arthroplasty over the last few decades, fracture identification related to THA.
patients continue to suffer from failed arthroplasties,
post-operative complications, and revisions. ML allows 2.1.3. Implant positioning
4
surgeons to integrate comorbidities, demographics, and Deep learning, a tool that requires less human intervention
pre-operative data into a model that can predict post- than traditional ML, also offers significant potential
operative outcomes on a patient-by-patient basis. 2 for automating implant positioning measurements on
19
There have been efforts to integrate ML into hip, knee, radiographs. Rouzrokh et al. developed a DL tool to
and shoulder arthroplasty. Several models are already measure acetabular inclination and version, demonstrating
5,6
in use, optimizing implant design, analyzing patient minimal discrepancies (mean differences of 1.35° for
comorbidities, and predicting restoration of function. inclination and 1.39° for anteversion) compared to expert
2
These models can synthesize the information provided human evaluations. In a separate study, the same team
by large data sets and apply it to individual patient cases, investigated femoral component subsidence, finding a
20
allowing them to serve as tools to augment the clinician’s mean difference of only 0.6 mm from expert assessments.
acumen. Therefore, the use of such predictive ML can These advancements hold promise for streamlining pre-
5-7
assist physicians in making informed decisions during all operative planning and intraoperative guidance in revision
steps of patient care, providing the potential to improve surgeries. Table 1 summarizes key studies examining the
outcomes after arthroplasty. 7,8 use of AI in implant positioning related to THA.
2. Hip arthroplasty 2.1.4. Disease identification

2.1. Radiographic imaging Recent advances in ML show significant promise in
diagnosing various orthopedic conditions affecting
2.1.1. Implant identification the hip. For hip dysplasia, AI software demonstrates
One promising area of ML in orthopedics is the tool’s ability good to excellent agreement with expert assessments,
21
to identify implants preoperatively. Multiple models have while significantly reducing diagnosis time by 80.4%.
been shown to accurately predict the ability of trained DCNNs have been successfully applied in diagnosing
22
models to discriminate between various implant models, hip osteoarthritis (OA), with Xue et al. pioneering this
both internally and externally, with current studies placing approach and achieving an impressive 95.0% sensitivity
the accuracy between 98% and 100%. 9-12 Given that 10% and 90.7% specificity. These results further validated
23
of implants are not identified before revision, the ability to by Üreten et al. demonstrated the ability of CNNs to
use ML to fill this gap may have substantial pre-operative rival the diagnostic accuracy of experienced physicians.
planning implications. When comparing the convolutional Furthermore, CNNs exhibited potential for osteoporosis
13
neural network’s (CNN) ability to identify implants to evaluation by directly analyzing bone mineral density
practicing surgeons, most surgeons outperformed the from radiographs, showing a strong correlation with
24
older models used in this study. However, the model saved dual-energy X-ray absorptiometry measurements. These
significant time in 17% of cases, hinting at a promising findings suggest that AI-assisted tools could streamline
solution that may address pre-operative implant planning the diagnostic process for hip-related pathologies, offering
in the near future. Table 1 summarizes key studies that valuable support to clinicians in orthopedic practice.
14
have explored the use of AI in implant identification related Table 1 outlines key studies exploring the use of AI in
to total hip arthroplasties (THA). disease identification related to THA.

Volume 2 Issue 2 (2025) 12 doi: 10.36922/aih.3278