Automated Alignment of Ultrasound Images for Pediatric Distal Forearm Fracture Diagnosis
Overview
This study introduces an automated pipeline to align point-of-care ultrasound (POCUS) images from dorsal and palmar views of the pediatric distal radius, enabling lateral X-ray-like visualization without radiation. The method uses segmentation-derived landmarks and anatomical constraints to optimize alignment, facilitating accurate fracture angulation and displacement assessment.
Background
Distal forearm fractures are the most common fractures in children, representing 29% of all pediatric fractures. Traditional diagnosis relies on X-rays, which expose children to radiation and require uncomfortable positioning. POCUS offers a radiation-free, quick, and child-friendly alternative but differs fundamentally from X-rays and lacks standardized methods for precise fracture characterization. Aligning dorsal and palmar POCUS images to create a lateral view could replace X-rays for fracture evaluation, but manual alignment is impractical in emergency settings. This study proposes an automated alignment approach to overcome these challenges.
Data Highlights
The alignment method uses three automatically extracted landmarks per POCUS image—physis position and two diaphysis points—derived from nnUnet-based bone segmentation. Three anatomical constraints guide the optimization: latitudinal alignment of physis points, parallel orientation of diaphysis vectors, and maintenance of bone width between diaphysis landmarks. Gradient descent with 10,000 iterations and a learning rate of 0.05 is employed to minimize the combined constraint function, achieving anatomically correct alignment of dorsal and palmar radius images.
Key Findings
POCUS images are acquired in six longitudinal planes, with dorsal and palmar views of the radius used for alignment.
Automated segmentation identifies three key landmarks per image critical for alignment: physis and two diaphysis points.
Three anatomical constraints (latitudinal, longitudinal orientation, and longitudinal position) effectively guide the alignment optimization.
The proposed pipeline uses gradient descent to iteratively align images, producing a lateral X-ray-like view without ulna superposition.
This approach enables precise quantification of fracture angulation and physeal displacement comparable to conventional lateral X-rays.
The method addresses the challenge of limited overlap and noisy POCUS images by leveraging anatomical knowledge and segmentation-derived landmarks.
Clinical Implications
The automated alignment of dorsal and palmar POCUS images can facilitate radiation-free, accurate diagnosis of distal forearm fractures in children, potentially eliminating the need for confirmatory X-rays. This technique supports precise assessment of fracture angulation and displacement, improving fracture management while enhancing patient comfort and safety. Implementation in emergency care could streamline workflows and reduce radiation exposure in pediatric fracture evaluation.
Conclusion
This pioneering automated alignment pipeline transforms POCUS imaging into a reliable, X-ray-independent diagnostic tool for pediatric distal forearm fractures, enabling detailed fracture characterization through anatomically guided image fusion. It holds promise for safer, faster, and more child-friendly fracture diagnosis.
