An artificial intelligence framework for universal landmark matching and morphometry in musculoskeletal radiography

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Accurate musculoskeletal morphometric measurements are vital but labor-intensive and prone to inter-reader variability.
• Existing AI solutions often require extensive annotated data and have narrow applications.
• A training-free AI framework using universal landmark matching is proposed to address these limitations.

Purpose of the Study:

• To present and validate a training-free artificial intelligence (AI) framework for automatic morphometric measurements in musculoskeletal radiography.
• To assess the framework's accuracy and generalizability across multiple anatomies and radiographic views.
• To compare the AI-derived measurements against manual annotations and radiologist performance.

Main Methods:

• A retrospective study analyzed 600 standard radiographs (foot, knee, shoulder) and 240 challenging cases with orthopedic implants.
• A pre-trained generalist dense-matching method transferred landmarks from reference radiographs to unseen images.
• Measurements were derived in a post-processing step and compared with manual measurements by two radiologists.

Main Results:

• Mean landmark matching error improved from 2.68 mm (1 reference) to 2.15 mm (40 references).
• Measurement accuracy ranged from 1.81° to 8.65°, approaching inter-reader agreement with more reference images.
• Performance on challenging cases was mixed, indicating specific limitations and strengths of the AI approach.

Conclusions:

• The anatomy-agnostic, training-free framework enables morphometry across multiple regions with performance comparable to inter-reader agreement.
• Challenging cases necessitate quality control and reference-set tuning for optimal deployment.
• The framework's minimal setup allows rapid adaptation to new anatomies and measurements, with clinically practical runtimes.