Applying machine learning methods to enable automatic customisation of knee replacement implants from CT data

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Artificial Intelligence

Background:

• Total knee replacement (TKR) is a common orthopedic procedure.
• Current implant design methods can be time-consuming and costly.
• Patient-specific implants may offer improved outcomes.

Purpose of the Study:

• To develop an automated pipeline for custom total knee replacement implant design.
• To leverage machine learning and computer-aided design for personalized implants.
• To assess the accuracy and feasibility of the automated CT-based design process.

Main Methods:

• Utilized machine learning (classification, object detection, segmentation) on CT scans (DICOM files).
• Employed statistical shape models to predict femur and tibia 3D surface models.
• Integrated computer-aided design (CAD) scripts for automated implant generation.
• Trained and validated the pipeline using data from 98 Korean Asian subjects.

Main Results:

• The pipeline demonstrated repeatable and highly accurate custom implant designs.
• Computational fitting showed excellent congruence with ground truth bone models.
• Performance was consistent across variations in subject sex, age, height, and knee side.

Conclusions:

• An automated, CT-based pipeline for custom TKR implant design is feasible.
• The system offers significant time and cost advantages over conventional methods.
• The framework is adaptable for customising other medical implants.