High accuracy in lower limb alignment analysis using convolutional neural networks, with improvements needed for...

Area of Science:

• Orthopedic surgery
• Medical imaging analysis
• Artificial intelligence in healthcare

Background:

• Long-leg standing radiographs (LSR) are standard for lower limb deformity assessment.
• Deep-learning convolutional neural networks (CNNs) can potentially improve radiological measurement accuracy and reproducibility.
• Automated tools may streamline preoperative planning for orthopedic procedures.

Purpose of the Study:

• To evaluate the measurement accuracy of an automated CNN-based planning tool (mediCAD® 7.0) for lower limb deformities.
• To compare CNN measurements against manual measurements by observers.
• To assess the tool's performance in patients undergoing total knee arthroplasty (TKA).

Main Methods:

• Retrospective analysis of 164 bilateral LSRs from patients with knee arthritis undergoing TKA.
• Independent analysis of alignment parameters by two observers and a CNN.
• Evaluation of accuracy using intraclass correlation coefficient (ICC), absolute deviations, limits of agreement (LoA), and root mean square error (RMSE).

Main Results:

• CNN demonstrated high consistency for leg length (ICC > 0.99) and overall lower limb alignment (mTFA ICC > 0.97).
• Mean absolute angular differences were low for overall alignment (mTFA 0.49-0.61°) but higher for specific joint angles (aMPFA 3.86-4.50°).
• Accuracy for specific joint angles (MPTA, mLDFA) varied, with greatest differences observed in TKA cases (ICC 0.22-0.85).

Conclusions:

• Excellent accuracy was achieved for overall alignment and leg length measurements compared to manual methods.
• Joint-level measurement accuracy needs enhancement, particularly in TKA patients, aligning with limitations of other algorithms.
• The automated tool significantly improves preoperative planning efficiency despite observed deviations.