Novel Technique for the Identification of Hip Implants Using Artificial Intelligence.

Area of Science:

• Orthopedic surgery
• Medical imaging
• Artificial intelligence

Background:

• Increasing total hip arthroplasty (THA) necessitates more revision THA.
• Accurate preoperative planning, including implant identification, is crucial for successful revision surgery.
• Existing AI methods for hip implant identification have limitations in dataset size, implant similarity, scalability, and AI expertise requirements.

Purpose of the Study:

• To develop a novel, scalable technique for generating large datasets of radiographically similar hip implants.
• To test the efficacy of a no-code machine learning solution for identifying these implants.
• To improve preoperative planning for revision total hip arthroplasty.

Main Methods:

• A convolutional neural network was trained and validated using a novel technique generating 27,020 computed tomography (CT)-derived projection images of 9 radiographically similar femoral implants.
• The technique involved superimposing 3D scanned implant models within CT pelvis volumes using computer-aided design and MATLAB.
• A separate set of 786 images was used for testing the model's performance.

Main Results:

• The machine learning model achieved a mean accuracy of 97.4% in discriminating between the 9 implant models.
• The model demonstrated high specificity (98.5%) and good sensitivity (88.4%) for implant identification.
• The no-code approach yielded meaningful results without requiring specialized AI expertise.

Conclusions:

• The developed hip implant detection technique accurately identifies 9 radiographically similar implants.
• This novel method enables the generation of large, scalable datasets, accommodating historic or obscure implants.
• The no-code machine learning model proves the feasibility of achieving significant results in implant identification, paving the way for further research and clinical application.