Quantifying 3D foot and ankle alignment using an AI-driven framework: a pilot study.

Area of Science:

• Orthopedics and Medical Imaging
• Artificial Intelligence in Healthcare
• Biomechanical Analysis

Background:

• Accurate foot and ankle alignment assessment is crucial for diagnosing deformities and guiding treatment.
• Traditional 2D radiographs have limitations in capturing the 3D complexity of foot and ankle structures.
• Weight-bearing CT (WBCT) offers a 3D view under physiological load, but manual landmark identification is time-consuming and variable.

Purpose of the Study:

• To present a novel AI framework for automating foot and ankle alignment assessment using deep learning landmark detection on WBCT images.
• To eliminate the need for manual segmentation and iterative mesh registration in WBCT analysis.
• To evaluate the accuracy of the AI framework in a clinically relevant population with foot deformities.

Main Methods:

• Developed and trained 3D U-Net models to predict 22 anatomical landmarks directly from WBCT images using heatmap predictions.
• Utilized a dataset of 74 orthopedic patients with conditions like pes cavus and planovalgus.
• Assessed mean absolute error for landmarks and angles via fivefold cross-validation.

Main Results:

• Mean absolute distance errors for landmark identification ranged from 1.00 mm to 1.88 mm.
• Automated clinical measurements derived from AI landmarks showed mean absolute errors between 0.91° for hindfoot angle and 2.90° for Böhler angle.
• The AI approach achieved accuracies comparable to manual inter-rater variability.

Conclusions:

• The heatmap-based AI approach enables automated foot and ankle alignment assessment from WBCT imaging with high accuracy.
• This AI-driven method offers a potentially valuable tool for evaluating foot and ankle morphology.
• Further validation with larger datasets is recommended to confirm clinical applicability.