Predicting CT-based coronal plane knee phenotype parameters using imageless navigation and machine learning.

Area of Science:

• Orthopedic surgery
• Biomechanical engineering
• Medical imaging analysis

Background:

• Coronal plane alignment of the knee (CPAK) is crucial for knee phenotypes, traditionally assessed with long leg radiographs.
• Alternative methods like CT, image-based, and imageless navigation are emerging for CPAK assessment.
• Machine learning (ML) offers data-driven approaches to improve accuracy in medical measurements.

Purpose of the Study:

• To evaluate the accuracy of imageless navigation in measuring CPAK parameters (JLO, aHKA).
• To compare imageless navigation measurements against CT-based assessments.
• To investigate the role of ML models and cartilage wear assumptions in enhancing accuracy.

Main Methods:

• Retrospective review of 152 total knee arthroplasties (TKAs) using imageless navigation and preoperative CT data.
• Measurement of MPTA and LDFA from both CT and imageless navigation data.
• Application of three cartilage wear assumptions to imageless navigation data, including ML-based models.
• Calculation of Mean Absolute Error (MAE) to quantify discrepancies between methods.

Main Results:

• ML-based wear assumptions yielded the lowest MAE for all CPAK parameters (≤1.2° for MPTA/LDFA, ≤1.8° for JLO/aHKA).
• This performance surpassed models with no wear correction (MAE 2.5° for aHKA) and generic wear assumptions (MAE 2.6° for aHKA).
• The study included 152 TKAs with an average patient age of 73 years and 61% female participants.

Conclusions:

• Imageless navigation, when integrated with ML models for cartilage wear prediction, accurately measures MPTA and LDFA with a mean error below 1.2° compared to CT.
• These findings support the use of imageless navigation as an effective tool for CPAK parameter measurement.
• Imageless navigation achieves results comparable to CT-based approaches for assessing coronal knee alignment.